support dirty restraint on vCPU
v16 cover support dirty restraint on vCPU
From: Hyman Huang(黄勇) <[email protected]>
v16
- rebase on master
- drop the unused typedef syntax in [PATCH v15 6/7]
- add the Reviewed-by and Acked-by tags by the way
v15
- rebase on master
- drop the 'init_time_ms' parameter in function vcpu_calculate_dirtyrate
- drop the 'setup' field in dirtylimit_state and call dirtylimit_process directly, which makes code cleaner.
- code clean in dirtylimit_adjust_throttle
- fix miss dirtylimit_state_unlock() in dirtylimit_process and dirtylimit_query_all
- add some comment
Please review. Thanks,
Regards
Yong
v14
- v13 sent by accident, resend patchset.
v13
- rebase on master
- passing NULL to kvm_dirty_ring_reap in commit
"refactor per-vcpu dirty ring reaping" to keep the logic unchanged. In other word, we still try the best to reap as much PFNs as possible if dirtylimit not in service.
- move the cpu list gen id changes into a separate patch.
- release the lock before sleep during dirty page rate calculation.
- move the dirty ring size fetch logic into a separate patch.
- drop the DIRTYLIMIT_LINEAR_ADJUSTMENT_WATERMARK MACRO .
- substitute bh with function pointer when implement dirtylimit.
- merge the dirtylimit_start/stop into dirtylimit_change.
- fix "cpu-index" parameter type with "int" to keep consistency.
- fix some syntax error in documents.
Please review. Thanks,
Yong
v12
- rebase on master
- add a new commmit to refactor per-vcpu dirty ring reaping, which can resolve the "vcpu miss the chances to sleep" problem
- remove the dirtylimit_thread and implemtment throttle in bottom half instead.
- let the dirty ring reaper thread keep sleeping when dirtylimit is in service
- introduce cpu_list_generation_id to identify cpu_list changing.
- keep taking the cpu_list_lock during dirty_stat_wait to prevent vcpu plug/unplug when calculating the dirty page rate
- move the dirtylimit global initializations out of dirtylimit_set_vcpu and do some code clean
- add DIRTYLIMIT_LINEAR_ADJUSTMENT_WATERMARK in case of oscillation when throttling
- remove the unmatched count field in dirtylimit_state
- add stub to fix build on non-x86
- refactor the documents
Thanks Peter and Markus for reviewing the previous versions, please review.
Thanks,
Yong
v11
- rebase on master
- add a commit " refactor dirty page rate calculation" so that dirty page rate limit can reuse the calculation logic.
- handle the cpu hotplug/unplug case in the dirty page rate calculation logic.
- modify the qmp commands according to Markus's advice.
- introduce a standalone file dirtylimit.c to implement dirty page rate limit
- check if dirty limit in service by dirtylimit_state pointer instead of global variable
- introduce dirtylimit_mutex to protect dirtylimit_state
- do some code clean and docs
See the commit for more detail, thanks Markus and Peter very mush for the code review and give the experienced and insightful advices, most modifications are based on these advices.
v10:
- rebase on master
- make the following modifications on patch [1/3]:
1. Make "dirtylimit-calc" thread joinable and join it after quitting.
2. Add finalize function to free dirtylimit_calc_state
3. Do some code clean work
- make the following modifications on patch [2/3]:
1. Remove the original implementation of throttle according to Peter's advice.
2. Introduce a negative feedback system and implement the throttle on all vcpu in one thread named "dirtylimit".
3. Simplify the algo when calculation the throttle_us_per_full:
increase/decrease linearly when there exists a wide difference between quota and current dirty page rate, increase/decrease a fixed time slice when the difference is narrow. This makes throttle responds faster and reach the quota smoothly.
4. Introduce a unfit_cnt in algo to make sure throttle really takes effect.
5. Set the max sleep time 99 times more than "ring_full_time_us".
6. Make "dirtylimit" thread joinable and join it after quitting.
- make the following modifications on patch [3/3]:
1. Remove the unplug cpu handling logic.
2. "query-vcpu-dirty-limit" only return dirtylimit information of vcpus that enable dirtylimit
3. Remove the "dirtylimit_setup" function
4. Trigger the dirtylimit and initialize the global state only when someone enable dirtylimit, and finalize it after the last dirtylimit be canceled.
5. Redefine the qmp command vcpu-dirty-limit/query-vcpu-dirty-limit:
enable/disable dirtylimit use a single command "vcpu-dirty-limit", to enable/disabled dirtylimit on specified vcpu only if "cpu-index" is specified, otherwise, all vcpu will be affected.
6. Redefine the hmp command vcpu_dirty_limit/info vcpu_dirty_limit
- other points about the code review:
1. "merge the code of calculation dirty page rate"
I think maybe it's not suitable to touch the 'calc-dirty-rate', because 'calc-dirty-rate' will stop sync log after calculating the dirtyrate and the 'dirtylimit-cal' will not untill the last dirtylimit be canceled, if we merge the GLOBAL_DIRTY_LIMIT into GLOBAL_DIRTY_DIRTYRATE, the two are interacted with each other.
2. The new implementaion of throttle algo enlightened by Peter responds faster and consume less cpu resource than the older, we make a impressed progress.
And there is a viewpoint may be discussed, it is that the new throttle logic is "passive", vcpu sleeps only after dirty ring, is full, unlike the "auto-converge" which will kick vcpu instead in a fixed slice time. If the vcpu is memory-write intensive and the ring size is large, it will produce dirty memory during the dirty ring full time and the throttle works not so good, it means the throttle depends on the dirty ring size.
I actually tested the new algo in two case:
case 1: dirty-ring-size: 4096, dirtyrate: 1170MB/s
result: minimum quota dirtyrate is 25MB/s or even less
minimum vcpu util is 6%
case 2: dirty-ring-size: 65536, dirtyrate: 1170MB/s
result: minimum quota dirtyrate is 256MB/s
minimum vcpu util is 24%
I post this just for discussion, i think this is not a big deal beacase if we set the dirty-ring-size to the maximum value(65536), we assume the server's bandwidth is capable of handling it.
3. I hard-code the acceptable deviation value to 25MB/s, see the macro DIRTYLIMIT_TOLERANCE_RANGE. I'm struggling to decide whether to let it configurable
4. Another point is the unplug cpu handle, current algo affects the unplugged vcpu, if we set dirty limit on it, we should fork 2 thread "dirtylimit" and "dirtylimit-calc" but do nothing, once the vcpu is hot-plugged, dirty limit works, i think the logic is ok but still there can be different advice.
- to let developers play with it easier, i post the hmp usage example:
(qemu) vcpu_dirty_limit -g on -1 500
[Please use 'info vcpu_dirty_limit' to query dirty limit for virtual CPU]
(qemu) info vcpu_dirty_limit
vcpu[0], limit rate 500 (MB/s), current rate 415 (MB/s) vcpu[1], limit rate 500 (MB/s), current rate 496 (MB/s) vcpu[2], limit rate 500 (MB/s), current rate 0 (MB/s) vcpu[3], limit rate 500 (MB/s), current rate 0 (MB/s)
(qemu) vcpu_dirty_limit -g off
[Please use 'info vcpu_dirty_limit' to query dirty limit for virtual CPU]
(qemu) info vcpu_dirty_limit
Dirty page limit not enabled!
(qemu) vcpu_dirty_limit on 0 300
[Please use 'info vcpu_dirty_limit' to query dirty limit for virtual CPU]
(qemu) vcpu_dirty_limit on 1 500
[Please use 'info vcpu_dirty_limit' to query dirty limit for virtual CPU]
(qemu) info vcpu_dirty_limit
vcpu[0], limit rate 300 (MB/s), current rate 342 (MB/s) vcpu[1], limit rate 500 (MB/s), current rate 485 (MB/s)
(qemu) vcpu_dirty_limit off 0
[Please use 'info vcpu_dirty_limit' to query dirty limit for virtual CPU]
(qemu) info vcpu_dirty_limit
vcpu[1], limit rate 500 (MB/s), current rate 528 (MB/s)
(qemu) vcpu_dirty_limit off 1
[Please use 'info vcpu_dirty_limit' to query dirty limit for virtual CPU]
(qemu) info vcpu_dirty_limit
Dirty page limit not enabled!
Thanks very much for the instructive algo suggestion given by Peter, the comment and other code reviews made by Markus.
Please review, thanks!
v9:
- rebase on master
- fix the meson directory change, keep it untouched.
v8:
- rebase on master
- polish the error message and remove the "unlikely" compilation syntax according to the advice given by Markus.
- keep the dirty tracking enabled during "dirtylimit-calc" lifecycle so that the overhead can be reduced according to the advice given by Peter.
- merge the "set/cancel" qmp commands into one named "vcpu-dirty-limit" and introduce qmp command "query-vcpu-dirty-limit" to query dirty limit information about virtual CPU, according to the advice given by Peter.
- check if vcpu index is valid and handle the unplug case before enabling, disabling dirty limit for virtual CPU.
- introduce hmp commands so developers can play with them easier, use "vcpu_dirty_limit" to enable dirty limit and "info vcpu_dirty_limit" to query.
The patch [2/3] has not been touched so far. Any corrections and suggetions are welcome.
Please review, thanks!
v7:
- rebase on master
- polish the comments and error message according to the advices given by Markus
- introduce dirtylimit_enabled function to pre-check if dirty page limit is enabled before canceling.
v6:
- rebase on master
- fix dirtylimit setup crash found by Markus
- polish the comments according to the advice given by Markus
- adjust the qemu qmp command tag to 7.0
v5:
- rebase on master
- adjust the throttle algorithm by removing the tuning in RESTRAINT_RATIO case so that dirty page rate could reachs the quota more quickly.
- fix percentage update in throttle iteration.
v4:
- rebase on master
- modify the following points according to the advice given by Markus
1. move the defination into migration.json
2. polish the comments of set-dirty-limit
3. do the syntax check and change dirty rate to dirty page rate
Thanks for the carefule reviews made by Markus.
Please review, thanks!
v3:
- rebase on master
- modify the following points according to the advice given by Markus
1. remove the DirtyRateQuotaVcpu and use its field as option directly
2. add comments to show details of what dirtylimit setup do
3. explain how to use dirtylimit in combination with existing qmp commands "calc-dirty-rate" and "query-dirty-rate" in documentation.
Thanks for the carefule reviews made by Markus.
Please review, thanks!
Hyman
v2:
- rebase on master
- modify the following points according to the advices given by Juan
1. rename dirtyrestraint to dirtylimit
2. implement the full lifecyle function of dirtylimit_calc, include dirtylimit_calc and dirtylimit_calc_quit
3. introduce 'quit' field in dirtylimit_calc_state to implement the dirtylimit_calc_quit
4. remove the ready_cond and ready_mtx since it may not be suitable
5. put the 'record_dirtypage' function code at the beggining of the file
6. remove the unnecesary return;
- other modifications has been made after code review
1. introduce 'bmap' and 'nr' field in dirtylimit_state to record the number of running thread forked by dirtylimit
2. stop the dirtyrate calculation thread if all the dirtylimit thread are stopped
3. do some renaming works
dirtyrate calulation thread -> dirtylimit-calc
dirtylimit thread -> dirtylimit-{cpu_index}
function name do_dirtyrestraint -> dirtylimit_check
qmp command dirty-restraint -> set-drity-limit
qmp command dirty-restraint-cancel -> cancel-dirty-limit
header file dirtyrestraint.h -> dirtylimit.hPlease review, thanks !
thanks for the accurate and timely advices given by Juan. we really appreciate it if corrections and suggetions about this patchset are proposed.
Best Regards !
Hyman
v1:
this patchset introduce a mechanism to impose dirty restraint on vCPU, aiming to keep the vCPU running in a certain dirtyrate given by user. dirty restraint on vCPU maybe an alternative method to implement convergence logic for live migration, which could improve guest memory performance during migration compared with traditional method in theory.
For the current live migration implementation, the convergence logic throttles all vCPUs of the VM, which has some side effects. -'read processes' on vCPU will be unnecessarily penalized
- throttle increase percentage step by step, which seems struggling to find the optimal throttle percentage when dirtyrate is high.
- hard to predict the remaining time of migration if the throttling percentage reachs 99%
to a certain extent, the dirty restraint machnism can fix these effects by throttling at vCPU granularity during migration.
the implementation is rather straightforward, we calculate vCPU dirtyrate via the Dirty Ring mechanism periodically as the commit 0e21bf246 "implement dirty-ring dirtyrate calculation" does, for vCPU that be specified to impose dirty restraint, we throttle it periodically as the auto-converge does, once after throttling, we compare the quota dirtyrate with current dirtyrate, if current dirtyrate is not under the quota, increase the throttling percentage until current dirtyrate is under the quota.
this patchset is the basis of implmenting a new auto-converge method for live migration, we introduce two qmp commands for impose/cancel the dirty restraint on specified vCPU, so it also can be an independent api to supply the upper app such as libvirt, which can use it to implement the convergence logic during live migration, supplemented with the qmp 'calc-dirty-rate' command or whatever.
we post this patchset for RFC and any corrections and suggetions about the implementation, api, throttleing algorithm or whatever are very appreciated!
Please review, thanks !
Best Regards !
Hyman Huang (7):
accel/kvm/kvm-all: Refactor per-vcpu dirty ring reaping
cpus: Introduce cpu_list_generation_id migration/dirtyrate: Refactor dirty page rate calculation softmmu/dirtylimit: Implement vCPU dirtyrate calculation periodically accel/kvm/kvm-all: Introduce kvm_dirty_ring_size function softmmu/dirtylimit: Implement virtual CPU throttle softmmu/dirtylimit: Implement dirty page rate limit
accel/kvm/kvm-all.c | 45 +++- accel/stubs/kvm-stub.c | 5 + cpus-common.c | 8 + hmp-commands-info.hx | 13 + hmp-commands.hx | 32 +++ include/exec/cpu-common.h | 1 + include/exec/memory.h | 5 +- include/hw/core/cpu.h | 6 + include/monitor/hmp.h | 3 + include/sysemu/dirtylimit.h | 37 +++ include/sysemu/dirtyrate.h | 28 +++ include/sysemu/kvm.h | 2 + migration/dirtyrate.c | 227 +++++++++++------ migration/dirtyrate.h | 7 +- qapi/migration.json | 80 ++++++ softmmu/dirtylimit.c | 602 ++++++++++++++++++++++++++++++++++++++++++++ softmmu/meson.build | 1 + softmmu/trace-events | 7 + 18 files changed, 1010 insertions(+), 99 deletions(-) create mode 100644 include/sysemu/dirtylimit.h create mode 100644 include/sysemu/dirtyrate.h create mode 100644 softmmu/dirtylimit.c
Signature · 1 line
1.8.3.1
PATCH v16 1/7 accel/kvm/kvm-all: Refactor per-vcpu dirty ring reaping
From: Hyman Huang(黄勇) <[email protected]>
Add a non-required argument 'CPUState' to kvm_dirty_ring_reap so that it can cover single vcpu dirty-ring-reaping scenario.
--- accel/kvm/kvm-all.c | 23 +++++++++++++---------- 1 file changed, 13 insertions(+), 10 deletions(-)
accel/kvm/kvm-all.c +13 -10
@@ -756,17 +756,20 @@ static uint32_t kvm_dirty_ring_reap_one(KVMState *s, CPUState *cpu) }/* Must be with slots_lock held */ -static uint64_t kvm_dirty_ring_reap_locked (KVMState *s) +static uint64_t kvm_dirty_ring_reap_locked (KVMState *s,CPUState * cpu) {int ret; -CPUState *cpu;uint64_t total =0 ;int64_t stamp; stamp =get_clock (); -CPU_FOREACH (cpu) { - total +=kvm_dirty_ring_reap_one (s, cpu); +if (cpu) { + total =kvm_dirty_ring_reap_one (s, cpu); + }else { +CPU_FOREACH (cpu) { + total +=kvm_dirty_ring_reap_one (s, cpu); + } }if (total) {
@@ -787,7 +790,7 @@ static uint64_t kvm_dirty_ring_reap_locked(KVMState *s)* Currently for simplicity, we must hold BQL before calling this. We can * consider to drop the BQL if we're clear with all the race conditions. */ -static uint64_t kvm_dirty_ring_reap (KVMState *s) +static uint64_t kvm_dirty_ring_reap (KVMState *s,CPUState *cpu) {uint64_t total;
@@ -807,7 +810,7 @@ static uint64_t kvm_dirty_ring_reap(KVMState *s)* reset below. */ kvm_slots_lock (); - total =kvm_dirty_ring_reap_locked (s); + total =kvm_dirty_ring_reap_locked (s, cpu);kvm_slots_unlock ();return total;
@@ -854,7 +857,7 @@ static void kvm_dirty_ring_flush(void)* vcpus out in a synchronous way. */ kvm_cpu_synchronize_kick_all (); -kvm_dirty_ring_reap (kvm_state); +kvm_dirty_ring_reap (kvm_state,NULL );trace_kvm_dirty_ring_flush (1 ); }
@@ -1398,7 +1401,7 @@ static void kvm_set_phys_mem(KVMMemoryListener *kml,* Not easy. Let's cross the fingers until it's fixed. */ if (kvm_state->kvm_dirty_ring_size) { -kvm_dirty_ring_reap_locked (kvm_state); +kvm_dirty_ring_reap_locked (kvm_state,NULL ); }else {kvm_slot_get_dirty_log (kvm_state, mem); }
@@ -1470,7 +1473,7 @@ static void *kvm_dirty_ring_reaper_thread(void *data) r->reaper_state =KVM_DIRTY_RING_REAPER_REAPING ;qemu_mutex_lock_iothread (); -kvm_dirty_ring_reap (s); +kvm_dirty_ring_reap (s,NULL );qemu_mutex_unlock_iothread (); r->reaper_iteration++;
@@ -2956,7 +2959,7 @@ int kvm_cpu_exec(CPUState *cpu)*/ trace_kvm_dirty_ring_full (cpu->cpu_index);qemu_mutex_lock_iothread (); -kvm_dirty_ring_reap (kvm_state); +kvm_dirty_ring_reap (kvm_state,NULL );qemu_mutex_unlock_iothread (); ret =0 ;break ;
Signature · 1 line
1.8.3.1
PATCH v16 2/7 cpus: Introduce cpu_list_generation_id
In reply to [email protected]From: Hyman Huang(黄勇) <[email protected]>
Introduce cpu_list_generation_id to track cpu list generation so that cpu hotplug/unplug can be detected during measurement of dirty page rate.
cpu_list_generation_id could be used to detect changes of cpu list, which is prepared for dirty page rate measurement.
--- cpus-common.c | 8 ++++++++ include/exec/cpu-common.h | 1 + 2 files changed, 9 insertions(+)
cpus-common.c +8 -0
@@ -73,6 +73,12 @@ static int cpu_get_free_index(void) }CPUTailQ cpus =QTAILQ_HEAD_INITIALIZER (cpus); +static unsigned int cpu_list_generation_id; + +unsigned int cpu_list_generation_id_get (void ) +{ +return cpu_list_generation_id; +}void cpu_list_add (CPUState *cpu) {
@@ -84,6 +90,7 @@ void cpu_list_add(CPUState *cpu)assert (!cpu_index_auto_assigned); }QTAILQ_INSERT_TAIL_RCU (&cpus, cpu, node); + cpu_list_generation_id++; }void cpu_list_remove (CPUState *cpu)
@@ -96,6 +103,7 @@ void cpu_list_remove(CPUState *cpu)QTAILQ_REMOVE_RCU (&cpus, cpu, node); cpu->cpu_index =UNASSIGNED_CPU_INDEX ; + cpu_list_generation_id++; }CPUState *qemu_get_cpu (int index)
include/exec/cpu-common.h +1 -0
@@ -20,6 +20,7 @@ extern intptr_t qemu_host_page_mask;void qemu_init_cpu_list (void );void cpu_list_lock (void );void cpu_list_unlock (void ); +unsigned int cpu_list_generation_id_get (void );void tcg_flush_softmmu_tlb (CPUState *cs);
Signature · 1 line
1.8.3.1
PATCH v16 3/7 migration/dirtyrate: Refactor dirty page rate calculation
In reply to [email protected]From: Hyman Huang(黄勇) <[email protected]>
abstract out dirty log change logic into functionglobal_dirty_log_change.
abstract out dirty page rate calculation logic via dirty-ring into function vcpu_calculate_dirtyrate.
abstract out mathematical dirty page rate calculation into do_calculate_dirtyrate, decouple it from DirtyStat.
rename set_sample_page_period to dirty_stat_wait, which is well-understood and will be reused in dirtylimit.
handle cpu hotplug/unplug scenario during measurement of dirty page rate.
export util functions outside migration.
--- include/sysemu/dirtyrate.h | 28 ++++++ migration/dirtyrate.c | 227 ++++++++++++++++++++++++++++----------------- migration/dirtyrate.h | 7 +- 3 files changed, 174 insertions(+), 88 deletions(-) create mode 100644 include/sysemu/dirtyrate.h
include/sysemu/dirtyrate.h new file +28 -0
@@ -0,0 +1,28 @@ +/* +* dirty page rate helper functions +* +* Copyright (c) 2022 CHINA TELECOM CO.,LTD. +* +* Authors: +* Hyman Huang(黄勇) <[email protected]> +* +* This work is licensed under the terms of the GNU GPL, version 2 or later. +* See the COPYING file in the top-level directory. +*/ + +#ifndef QEMU_DIRTYRATE_H +#define QEMU_DIRTYRATE_H + +typedef struct VcpuStat { +int nvcpu;/* number of vcpu */ +DirtyRateVcpu *rates;/* array of dirty rate for each vcpu */ +}VcpuStat ; + +int64_t vcpu_calculate_dirtyrate (int64_t calc_time_ms, +VcpuStat *stat, +unsigned int flag, +bool one_shot); + +void global_dirty_log_change (unsigned int flag, +bool start); +#endif
migration/dirtyrate.c +144 -83
290 lines not on this page · raw message
migration/dirtyrate.h +2 -5
@@ -13,6 +13,8 @@#ifndef QEMU_MIGRATION_DIRTYRATE_H #define QEMU_MIGRATION_DIRTYRATE_H +#include "sysemu/dirtyrate.h" +/* * Sample 512 pages per GB as default. */
@@ -65,11 +67,6 @@ typedef struct SampleVMStat {uint64_t total_block_mem_MB;/* size of total sampled pages in MB */ } SampleVMStat; -typedef struct VcpuStat { -int nvcpu;/* number of vcpu */ -DirtyRateVcpu *rates;/* array of dirty rate for each vcpu */ -}VcpuStat ; -/* * Store calculation statistics for each measure. */
Signature · 1 line
1.8.3.1
PATCH v16 4/7 softmmu/dirtylimit: Implement vCPU dirtyrate calculation periodically
In reply to [email protected]From: Hyman Huang(黄勇) <[email protected]>
Introduce the third method GLOBAL_DIRTY_LIMIT of dirty tracking for calculate dirtyrate periodly for dirty page rate limit.
Add dirtylimit.c to implement dirtyrate calculation periodly, which will be used for dirty page rate limit.
Add dirtylimit.h to export util functions for dirty page rate limit implementation.
--- include/exec/memory.h | 5 +- include/sysemu/dirtylimit.h | 22 +++++++++ softmmu/dirtylimit.c | 116 ++++++++++++++++++++++++++++++++++++++++++++ softmmu/meson.build | 1 + 4 files changed, 143 insertions(+), 1 deletion(-) create mode 100644 include/sysemu/dirtylimit.h create mode 100644 softmmu/dirtylimit.c
include/exec/memory.h +4 -1
@@ -69,7 +69,10 @@ static inline void fuzz_dma_read_cb(size_t addr,/* Dirty tracking enabled because measuring dirty rate */ #define GLOBAL_DIRTY_DIRTY_RATE (1U << 1) -#define GLOBAL_DIRTY_MASK (0x3) +/* Dirty tracking enabled because dirty limit */ +#define GLOBAL_DIRTY_LIMIT (1U << 2) + +#define GLOBAL_DIRTY_MASK (0x7)extern unsigned int global_dirty_tracking;
include/sysemu/dirtylimit.h new file +22 -0
@@ -0,0 +1,22 @@ +/* +* Dirty page rate limit common functions +* +* Copyright (c) 2022 CHINA TELECOM CO.,LTD. +* +* Authors: +* Hyman Huang(黄勇) <[email protected]> +* +* This work is licensed under the terms of the GNU GPL, version 2 or later. +* See the COPYING file in the top-level directory. +*/ +#ifndef QEMU_DIRTYRLIMIT_H +#define QEMU_DIRTYRLIMIT_H + +#define DIRTYLIMIT_CALC_TIME_MS 1000/* 1000ms */ + +int64_t vcpu_dirty_rate_get (int cpu_index); +void vcpu_dirty_rate_stat_start (void ); +void vcpu_dirty_rate_stat_stop (void ); +void vcpu_dirty_rate_stat_initialize (void ); +void vcpu_dirty_rate_stat_finalize (void ); +#endif
softmmu/dirtylimit.c new file +116 -0
116 lines not on this page · raw message
softmmu/meson.build +1 -0
@@ -15,6 +15,7 @@ specific_ss.add(when: 'CONFIG_SOFTMMU', if_true: [files( 'vl.c', 'cpu-timers.c', 'runstate-action.c', + 'dirtylimit.c', )]) specific_ss.add(when: ['CONFIG_SOFTMMU', 'CONFIG_TCG'], if_true: [files(
Signature · 1 line
1.8.3.1
PATCH v16 5/7 accel/kvm/kvm-all: Introduce kvm_dirty_ring_size function
In reply to [email protected]From: Hyman Huang(黄勇) <[email protected]>
Introduce kvm_dirty_ring_size util function to help calculate dirty ring ful time.
--- accel/kvm/kvm-all.c | 5 +++++ accel/stubs/kvm-stub.c | 5 +++++ include/sysemu/kvm.h | 2 ++ 3 files changed, 12 insertions(+)
accel/kvm/kvm-all.c +5 -0
@@ -2312,6 +2312,11 @@ bool kvm_dirty_ring_enabled(void)return kvm_state->kvm_dirty_ring_size ? true : false; } +uint32_t kvm_dirty_ring_size (void ) +{ +return kvm_state->kvm_dirty_ring_size; +} +static int kvm_init (MachineState *ms) {MachineClass *mc =MACHINE_GET_CLASS (ms);
accel/stubs/kvm-stub.c +5 -0
@@ -152,4 +152,9 @@ bool kvm_dirty_ring_enabled(void) {return false; } + +uint32_t kvm_dirty_ring_size (void ) +{ +return 0 ; +}#endif
include/sysemu/kvm.h +2 -0
@@ -563,4 +563,6 @@ bool kvm_cpu_check_are_resettable(void);bool kvm_arch_cpu_check_are_resettable (void );bool kvm_dirty_ring_enabled (void ); + +uint32_t kvm_dirty_ring_size (void );#endif
Signature · 1 line
1.8.3.1
PATCH v16 6/7 softmmu/dirtylimit: Implement virtual CPU throttle
In reply to [email protected]From: Hyman Huang(黄勇) <[email protected]>
Setup a negative feedback system when vCPU thread handling KVM_EXIT_DIRTY_RING_FULL exit by introducing throttle_us_per_full field in struct CPUState. Sleep throttle_us_per_full microseconds to throttle vCPU if dirtylimit is in service.
--- accel/kvm/kvm-all.c | 19 ++- include/hw/core/cpu.h | 6 + include/sysemu/dirtylimit.h | 15 +++ softmmu/dirtylimit.c | 291 ++++++++++++++++++++++++++++++++++++++++++++ softmmu/trace-events | 7 ++ 5 files changed, 337 insertions(+), 1 deletion(-)
accel/kvm/kvm-all.c +18 -1
@@ -45,6 +45,7 @@#include "qemu/guest-random.h" #include "sysemu/hw_accel.h" #include "kvm-cpus.h" +#include "sysemu/dirtylimit.h" #include "hw/boards.h"
@@ -476,6 +477,7 @@ int kvm_init_vcpu(CPUState *cpu, Error **errp) cpu->kvm_state = s; cpu->vcpu_dirty = true; cpu->dirty_pages =0 ; + cpu->throttle_us_per_full =0 ; mmap_size =kvm_ioctl (s,KVM_GET_VCPU_MMAP_SIZE ,0 );if (mmap_size <0 ) {
@@ -1469,6 +1471,11 @@ static void *kvm_dirty_ring_reaper_thread(void *data)*/ sleep (1 ); +/* keep sleeping so that dirtylimit not be interfered by reaper */ +if (dirtylimit_in_service ()) { +continue ; + } +trace_kvm_dirty_ring_reaper ("wakeup" ); r->reaper_state =KVM_DIRTY_RING_REAPER_REAPING ;
@@ -2964,8 +2971,18 @@ int kvm_cpu_exec(CPUState *cpu)*/ trace_kvm_dirty_ring_full (cpu->cpu_index);qemu_mutex_lock_iothread (); -kvm_dirty_ring_reap (kvm_state,NULL ); +/* We throttle vCPU by making it sleep once it exit from kernel +* due to dirty ring full. In the dirtylimit scenario, reaping +* all vCPUs after a single vCPU dirty ring get full result in +* the miss of sleep, so just reap the ring-fulled vCPU. +*/ +if (dirtylimit_in_service ()) { +kvm_dirty_ring_reap (kvm_state, cpu); + }else { +kvm_dirty_ring_reap (kvm_state,NULL ); + }qemu_mutex_unlock_iothread (); +dirtylimit_vcpu_execute (cpu); ret =0 ;break ;case KVM_EXIT_SYSTEM_EVENT :
include/hw/core/cpu.h +6 -0
@@ -411,6 +411,12 @@ struct CPUState {*/ bool throttle_thread_scheduled; +/* +* Sleep throttle_us_per_full microseconds once dirty ring is full +* if dirty page rate limit is enabled. +*/ +int64_t throttle_us_per_full; +bool ignore_memory_transaction_failures;/* Used for user-only emulation of prctl(PR_SET_UNALIGN). */
include/sysemu/dirtylimit.h +15 -0
@@ -19,4 +19,19 @@ void vcpu_dirty_rate_stat_start(void);void vcpu_dirty_rate_stat_stop (void );void vcpu_dirty_rate_stat_initialize (void );void vcpu_dirty_rate_stat_finalize (void ); + +void dirtylimit_state_lock (void ); +void dirtylimit_state_unlock (void ); +void dirtylimit_state_initialize (void ); +void dirtylimit_state_finalize (void ); +bool dirtylimit_in_service (void ); +bool dirtylimit_vcpu_index_valid (int cpu_index); +void dirtylimit_process (void ); +void dirtylimit_change (bool start); +void dirtylimit_set_vcpu (int cpu_index, +uint64_t quota, +bool enable); +void dirtylimit_set_all (uint64_t quota, +bool enable); +void dirtylimit_vcpu_execute (CPUState *cpu);#endif
softmmu/dirtylimit.c +291 -0
321 lines not on this page · raw message
softmmu/trace-events +7 -0
@@ -31,3 +31,10 @@ runstate_set(int current_state, const char *current_state_str, int new_state, co system_wakeup_request(int reason) "reason=%d" qemu_system_shutdown_request(int reason) "reason=%d" qemu_system_powerdown_request(void) "" + +#dirtylimit.c +dirtylimit_state_initialize(int max_cpus) "dirtylimit state initialize: max cpus %d" +dirtylimit_state_finalize(void) +dirtylimit_throttle_pct(int cpu_index, uint64_t pct, int64_t time_us) "CPU[%d] throttle percent: %" PRIu64 ", throttle adjust time %"PRIi64 " us" +dirtylimit_set_vcpu(int cpu_index, uint64_t quota) "CPU[%d] set dirty page rate limit %"PRIu64 +dirtylimit_vcpu_execute(int cpu_index, int64_t sleep_time_us) "CPU[%d] sleep %"PRIi64 " us"
Signature · 1 line
1.8.3.1
PATCH v16 7/7 softmmu/dirtylimit: Implement dirty page rate limit
In reply to [email protected]From: Hyman Huang(黄勇) <[email protected]>
Implement dirtyrate calculation periodically basing on dirty-ring and throttle virtual CPU until it reachs the quota dirty page rate given by user.
Introduce qmp commands "set-vcpu-dirty-limit", "cancel-vcpu-dirty-limit", "query-vcpu-dirty-limit" to enable, disable, query dirty page limit for virtual CPU.
Meanwhile, introduce corresponding hmp commands "set_vcpu_dirty_limit", "cancel_vcpu_dirty_limit", "info vcpu_dirty_limit" so the feature can be more usable.
--- hmp-commands-info.hx | 13 ++++ hmp-commands.hx | 32 +++++++++ include/monitor/hmp.h | 3 + qapi/migration.json | 80 +++++++++++++++++++++ softmmu/dirtylimit.c | 195 ++++++++++++++++++++++++++++++++++++++++++++++++++ 5 files changed, 323 insertions(+)
hmp-commands-info.hx +13 -0
@@ -865,6 +865,19 @@ SRST Display the vcpu dirty rate information. ERST + { + .name = "vcpu_dirty_limit", + .args_type = "", + .params = "", + .help = "show dirty page limit information of all vCPU", + .cmd = hmp_info_vcpu_dirty_limit, + }, + +SRST + ``info vcpu_dirty_limit`` + Display the vcpu dirty page limit information. +ERST + #if defined(TARGET_I386) { .name = "sgx",
hmp-commands.hx +32 -0
@@ -1744,3 +1744,35 @@ ERST "\n\t\t\t -b to specify dirty bitmap as method of calculation)", .cmd = hmp_calc_dirty_rate, }, + +SRST +``set_vcpu_dirty_limit`` + Set dirty page rate limit on virtual CPU, the information about all the + virtual CPU dirty limit status can be observed with ``info vcpu_dirty_limit`` + command. +ERST + + { + .name = "set_vcpu_dirty_limit", + .args_type = "dirty_rate:l,cpu_index:l?", + .params = "dirty_rate [cpu_index]", + .help = "set dirty page rate limit, use cpu_index to set limit" + "\n\t\t\t\t\t on a specified virtual cpu", + .cmd = hmp_set_vcpu_dirty_limit, + }, + +SRST +``cancel_vcpu_dirty_limit`` + Cancel dirty page rate limit on virtual CPU, the information about all the + virtual CPU dirty limit status can be observed with ``info vcpu_dirty_limit`` + command. +ERST + + { + .name = "cancel_vcpu_dirty_limit", + .args_type = "cpu_index:l?", + .params = "[cpu_index]", + .help = "cancel dirty page rate limit, use cpu_index to cancel" + "\n\t\t\t\t\t limit on a specified virtual cpu", + .cmd = hmp_cancel_vcpu_dirty_limit, + },
include/monitor/hmp.h +3 -0
@@ -131,6 +131,9 @@ void hmp_replay_delete_break(Monitor *mon, const QDict *qdict);void hmp_replay_seek (Monitor *mon,const QDict *qdict);void hmp_info_dirty_rate (Monitor *mon,const QDict *qdict);void hmp_calc_dirty_rate (Monitor *mon,const QDict *qdict); +void hmp_set_vcpu_dirty_limit (Monitor *mon,const QDict *qdict); +void hmp_cancel_vcpu_dirty_limit (Monitor *mon,const QDict *qdict); +void hmp_info_vcpu_dirty_limit (Monitor *mon,const QDict *qdict);void hmp_human_readable_text_helper (Monitor *mon,HumanReadableText *(*qmp_handler)(Error **));
qapi/migration.json +80 -0
86 lines not on this page · raw message
softmmu/dirtylimit.c +195 -0
206 lines not on this page · raw message
Signature · 1 line
1.8.3.1
Dr. David Alan Gilbert
AppliedQuoted 1 line · * [email protected] ([email protected]) wrote:
From: Hyman Huang(黄勇) <[email protected]>
Queued via my migration/hmp/etc tree
Quoted 399 lines
v16
- rebase on master
- drop the unused typedef syntax in [PATCH v15 6/7]
- add the Reviewed-by and Acked-by tags by the wayv15
- rebase on master
- drop the 'init_time_ms' parameter in functionvcpu_calculate_dirtyrate
- drop the 'setup' field indirtylimit_stateand calldirtylimit_processdirectly, which makes code cleaner.
- code clean indirtylimit_adjust_throttle
- fix missdirtylimit_state_unlock()indirtylimit_processanddirtylimit_query_all
- add some commentPlease review. Thanks,
Regards
Yongv14
- v13 sent by accident, resend patchset.v13
- rebase on master
- passing NULL tokvm_dirty_ring_reapin commit
"refactor per-vcpu dirty ring reaping" to keep the logic unchanged. In other word, we still try the best to reap as much PFNs as possible if dirtylimit not in service.
- move the cpu list gen id changes into a separate patch.
- release the lock before sleep during dirty page rate calculation.
- move the dirty ring size fetch logic into a separate patch.
- drop theDIRTYLIMIT_LINEAR_ADJUSTMENT_WATERMARKMACRO .
- substitute bh with function pointer when implement dirtylimit.
- merge the dirtylimit_start/stop intodirtylimit_change.
- fix "cpu-index" parameter type with "int" to keep consistency.
- fix some syntax error in documents.Please review. Thanks,
Yong
v12
- rebase on master
- add a new commmit to refactor per-vcpu dirty ring reaping, which can resolve the "vcpu miss the chances to sleep" problem
- remove thedirtylimit_threadand implemtment throttle in bottom half instead.
- let the dirty ring reaper thread keep sleeping when dirtylimit is in service
- introducecpu_list_generation_idto identifycpu_listchanging.
- keep taking thecpu_list_lockduringdirty_stat_waitto prevent vcpu plug/unplug when calculating the dirty page rate
- move the dirtylimit global initializations out ofdirtylimit_set_vcpuand do some code clean
- addDIRTYLIMIT_LINEAR_ADJUSTMENT_WATERMARKin case of oscillation when throttling
- remove the unmatched count field indirtylimit_state
- add stub to fix build on non-x86
- refactor the documentsThanks Peter and Markus for reviewing the previous versions, please review.
Thanks,
Yongv11
- rebase on master
- add a commit " refactor dirty page rate calculation" so that dirty page rate limit can reuse the calculation logic.
- handle the cpu hotplug/unplug case in the dirty page rate calculation logic.
- modify the qmp commands according to Markus's advice.
- introduce a standalone filedirtylimit.cto implement dirty page rate limit
- check if dirty limit in service bydirtylimit_statepointer instead of global variable
- introducedirtylimit_mutexto protectdirtylimit_state
- do some code clean and docsSee the commit for more detail, thanks Markus and Peter very mush for the code review and give the experienced and insightful advices, most modifications are based on these advices.
v10:
- rebase on master
- make the following modifications on patch [1/3]:
1. Make "dirtylimit-calc" thread joinable and join it after quitting.2. Add finalize function to free
dirtylimit_calc_state3. Do some code clean work
- make the following modifications on patch [2/3]:
1. Remove the original implementation of throttle according to Peter's advice.2. Introduce a negative feedback system and implement the throttle on all vcpu in one thread named "dirtylimit".
3. Simplify the algo when calculation the
throttle_us_per_full:
increase/decrease linearly when there exists a wide difference between quota and current dirty page rate, increase/decrease a fixed time slice when the difference is narrow. This makes throttle responds faster and reach the quota smoothly.4. Introduce a
unfit_cntin algo to make sure throttle really takes effect.5. Set the max sleep time 99 times more than "
ring_full_time_us".6. Make "dirtylimit" thread joinable and join it after quitting.
- make the following modifications on patch [3/3]:
1. Remove the unplug cpu handling logic.2. "query-vcpu-dirty-limit" only return dirtylimit information of vcpus that enable dirtylimit
3. Remove the "
dirtylimit_setup" function4. Trigger the dirtylimit and initialize the global state only when someone enable dirtylimit, and finalize it after the last dirtylimit be canceled.
5. Redefine the qmp command vcpu-dirty-limit/query-vcpu-dirty-limit:
enable/disable dirtylimit use a single command "vcpu-dirty-limit", to enable/disabled dirtylimit on specified vcpu only if "cpu-index" is specified, otherwise, all vcpu will be affected.6. Redefine the hmp command vcpu_dirty_limit/info
vcpu_dirty_limit- other points about the code review:
1. "merge the code of calculation dirty page rate"
I think maybe it's not suitable to touch the 'calc-dirty-rate', because 'calc-dirty-rate' will stop sync log after calculating the dirtyrate and the 'dirtylimit-cal' will not untill the last dirtylimit be canceled, if we merge theGLOBAL_DIRTY_LIMITintoGLOBAL_DIRTY_DIRTYRATE, the two are interacted with each other.2. The new implementaion of throttle algo enlightened by Peter responds faster and consume less cpu resource than the older, we make a impressed progress.
And there is a viewpoint may be discussed, it is that the new throttle logic is "passive", vcpu sleeps only after dirty ring, is full, unlike the "auto-converge" which will kick vcpu instead in a fixed slice time. If the vcpu is memory-write intensive and the ring size is large, it will produce dirty memory during the dirty ring full time and the throttle works not so good, it means the throttle depends on the dirty ring size.
I actually tested the new algo in two case:
case 1: dirty-ring-size: 4096, dirtyrate: 1170MB/s
result: minimum quota dirtyrate is 25MB/s or even less
minimum vcpu util is 6%case 2: dirty-ring-size: 65536, dirtyrate: 1170MB/s
result: minimum quota dirtyrate is 256MB/s
minimum vcpu util is 24%I post this just for discussion, i think this is not a big deal beacase if we set the dirty-ring-size to the maximum value(65536), we assume the server's bandwidth is capable of handling it.
3. I hard-code the acceptable deviation value to 25MB/s, see the macro
DIRTYLIMIT_TOLERANCE_RANGE. I'm struggling to decide whether to let it configurable4. Another point is the unplug cpu handle, current algo affects the unplugged vcpu, if we set dirty limit on it, we should fork 2 thread "dirtylimit" and "dirtylimit-calc" but do nothing, once the vcpu is hot-plugged, dirty limit works, i think the logic is ok but still there can be different advice.
- to let developers play with it easier, i post the hmp usage example:
(qemu)vcpu_dirty_limit-g on -1 500
[Please use 'infovcpu_dirty_limit' to query dirty limit for virtual CPU](qemu) info
vcpu_dirty_limit
vcpu[0], limit rate 500 (MB/s), current rate 415 (MB/s) vcpu[1], limit rate 500 (MB/s), current rate 496 (MB/s) vcpu[2], limit rate 500 (MB/s), current rate 0 (MB/s) vcpu[3], limit rate 500 (MB/s), current rate 0 (MB/s)
(qemu)vcpu_dirty_limit-g off
[Please use 'infovcpu_dirty_limit' to query dirty limit for virtual CPU](qemu) info
vcpu_dirty_limit
Dirty page limit not enabled!(qemu)
vcpu_dirty_limiton 0 300
[Please use 'infovcpu_dirty_limit' to query dirty limit for virtual CPU](qemu)
vcpu_dirty_limiton 1 500
[Please use 'infovcpu_dirty_limit' to query dirty limit for virtual CPU](qemu) info
vcpu_dirty_limit
vcpu[0], limit rate 300 (MB/s), current rate 342 (MB/s) vcpu[1], limit rate 500 (MB/s), current rate 485 (MB/s)(qemu)
vcpu_dirty_limitoff 0
[Please use 'infovcpu_dirty_limit' to query dirty limit for virtual CPU](qemu) info
vcpu_dirty_limit
vcpu[1], limit rate 500 (MB/s), current rate 528 (MB/s)(qemu)
vcpu_dirty_limitoff 1
[Please use 'infovcpu_dirty_limit' to query dirty limit for virtual CPU](qemu) info
vcpu_dirty_limit
Dirty page limit not enabled!Thanks very much for the instructive algo suggestion given by Peter, the comment and other code reviews made by Markus.
Please review, thanks!
v9:
- rebase on master
- fix the meson directory change, keep it untouched.v8:
- rebase on master
- polish the error message and remove the "unlikely" compilation syntax according to the advice given by Markus.
- keep the dirty tracking enabled during "dirtylimit-calc" lifecycle so that the overhead can be reduced according to the advice given by Peter.
- merge the "set/cancel" qmp commands into one named "vcpu-dirty-limit" and introduce qmp command "query-vcpu-dirty-limit" to query dirty limit information about virtual CPU, according to the advice given by Peter.
- check if vcpu index is valid and handle the unplug case before enabling, disabling dirty limit for virtual CPU.
- introduce hmp commands so developers can play with them easier, use "vcpu_dirty_limit" to enable dirty limit and "infovcpu_dirty_limit" to query.The patch [2/3] has not been touched so far. Any corrections and suggetions are welcome.
Please review, thanks!
v7:
- rebase on master
- polish the comments and error message according to the advices given by Markus
- introducedirtylimit_enabledfunction to pre-check if dirty page limit is enabled before canceling.v6:
- rebase on master
- fix dirtylimit setup crash found by Markus
- polish the comments according to the advice given by Markus
- adjust the qemu qmp command tag to 7.0v5:
- rebase on master
- adjust the throttle algorithm by removing the tuning inRESTRAINT_RATIOcase so that dirty page rate could reachs the quota more quickly.
- fix percentage update in throttle iteration.v4:
- rebase on master
- modify the following points according to the advice given by Markus
1. move the defination into migration.json
2. polish the comments of set-dirty-limit
3. do the syntax check and change dirty rate to dirty page rateThanks for the carefule reviews made by Markus.
Please review, thanks!
v3:
- rebase on master
- modify the following points according to the advice given by Markus
1. remove the DirtyRateQuotaVcpu and use its field as option directly
2. add comments to show details of what dirtylimit setup do
3. explain how to use dirtylimit in combination with existing qmp commands "calc-dirty-rate" and "query-dirty-rate" in documentation.Thanks for the carefule reviews made by Markus.
Please review, thanks!
Hyman
v2:
- rebase on master
- modify the following points according to the advices given by Juan
1. rename dirtyrestraint to dirtylimit
2. implement the full lifecyle function ofdirtylimit_calc, includedirtylimit_calcanddirtylimit_calc_quit
3. introduce 'quit' field indirtylimit_calc_stateto implement thedirtylimit_calc_quit
4. remove theready_condandready_mtxsince it may not be suitable
5. put the 'record_dirtypage' function code at the beggining of the file
6. remove the unnecesary return;
- other modifications has been made after code review
1. introduce 'bmap' and 'nr' field indirtylimit_stateto record the number of running thread forked by dirtylimit
2. stop the dirtyrate calculation thread if all the dirtylimit thread are stopped
3. do some renaming worksdirtyrate calulation thread -> dirtylimit-calc dirtylimit thread -> dirtylimit-{cpu_index} function name do_dirtyrestraint -> dirtylimit_check qmp command dirty-restraint -> set-drity-limit qmp command dirty-restraint-cancel -> cancel-dirty-limit header file dirtyrestraint.h -> dirtylimit.hPlease review, thanks !
thanks for the accurate and timely advices given by Juan. we really appreciate it if corrections and suggetions about this patchset are proposed.
Best Regards !
Hyman
v1:
this patchset introduce a mechanism to impose dirty restraint on vCPU, aiming to keep the vCPU running in a certain dirtyrate given by user. dirty restraint on vCPU maybe an alternative method to implement convergence logic for live migration, which could improve guest memory performance during migration compared with traditional method in theory.For the current live migration implementation, the convergence logic throttles all vCPUs of the VM, which has some side effects. -'read processes' on vCPU will be unnecessarily penalized
- throttle increase percentage step by step, which seems struggling to find the optimal throttle percentage when dirtyrate is high.
- hard to predict the remaining time of migration if the throttling percentage reachs 99%to a certain extent, the dirty restraint machnism can fix these effects by throttling at vCPU granularity during migration.
the implementation is rather straightforward, we calculate vCPU dirtyrate via the Dirty Ring mechanism periodically as the commit
0e21bf246"implement dirty-ring dirtyrate calculation" does, for vCPU that be specified to impose dirty restraint, we throttle it periodically as the auto-converge does, once after throttling, we compare the quota dirtyrate with current dirtyrate, if current dirtyrate is not under the quota, increase the throttling percentage until current dirtyrate is under the quota.this patchset is the basis of implmenting a new auto-converge method for live migration, we introduce two qmp commands for impose/cancel the dirty restraint on specified vCPU, so it also can be an independent api to supply the upper app such as libvirt, which can use it to implement the convergence logic during live migration, supplemented with the qmp 'calc-dirty-rate' command or whatever.
we post this patchset for RFC and any corrections and suggetions about the implementation, api, throttleing algorithm or whatever are very appreciated!
Please review, thanks !
Best Regards !
Hyman Huang (7):
accel/kvm/kvm-all: Refactor per-vcpu dirty ring reaping
cpus: Introducecpu_list_generation_idmigration/dirtyrate: Refactor dirty page rate calculation softmmu/dirtylimit: Implement vCPU dirtyrate calculation periodically accel/kvm/kvm-all: Introducekvm_dirty_ring_sizefunction softmmu/dirtylimit: Implement virtual CPU throttle softmmu/dirtylimit: Implement dirty page rate limitaccel/kvm/kvm-all.c | 45 +++- accel/stubs/kvm-stub.c | 5 + cpus-common.c | 8 + hmp-commands-info.hx | 13 + hmp-commands.hx | 32 +++ include/exec/cpu-common.h | 1 + include/exec/memory.h | 5 +- include/hw/core/cpu.h | 6 + include/monitor/hmp.h | 3 + include/sysemu/dirtylimit.h | 37 +++ include/sysemu/dirtyrate.h | 28 +++ include/sysemu/kvm.h | 2 + migration/dirtyrate.c | 227 +++++++++++------ migration/dirtyrate.h | 7 +- qapi/migration.json | 80 ++++++ softmmu/dirtylimit.c | 602 ++++++++++++++++++++++++++++++++++++++++++++ softmmu/meson.build | 1 + softmmu/trace-events | 7 + 18 files changed, 1010 insertions(+), 99 deletions(-) create mode 100644 include/sysemu/dirtylimit.h create mode 100644 include/sysemu/dirtyrate.h create mode 100644 softmmu/dirtylimit.cSignature · 1 line
1.8.3.1
Signature · 1 line
Dr. David Alan Gilbert / [email protected] / Manchester, UK
Quoted 4 lines · * Dr. David Alan Gilbert ([email protected]) wrote:
Quoted 1 line · * [email protected] ([email protected]) wrote:
From: Hyman Huang(黄勇) <[email protected]>
Queued via my migration/hmp/etc tree
Hi,
Unfortunately I've had to unqueue this - it breaks the
qmp-cmd-test:
# starting QEMU: exec ./x86_64-softmmu/qemu-system-x86_64 -qtest unix:/tmp/qtest-142136.sock -qtest-log /dev/fd/2 -chardev socket,path=/tmp/qtest-142136.qmp,id=char0 -mon chardev=char0,mode=control -display none -nodefaults -machine none -accel qtest
[I 1646239093.713627] OPENED
[R +0.000190] endianness
[S +0.000196] OK little
{"QMP": {"version": {"qemu": {"micro": 50, "minor": 2, "major": 6}, "package": "v6.2.0-1867-g817703d65a"}, "capabilities": ["oob"]}}{"execute": "qmp_capabilities"}{"return": {}}{"execute": "query-vcpu-dirty-limit"}
{"error": {"class": "GenericError", "desc": "dirty page limit not enabled"}}** ERROR:../tests/qtest/qmp-cmd-test.c:84:test_query: assertion failed: (qdict_haskey(resp, "return")) Bail out! ERROR:../tests/qtest/qmp-cmd-test.c:84:test_query: assertion failed: (qdict_haskey(resp, "return")) [I +0.195433] CLOSED
Aborted (core dumped)
qmp-cmd-test tries to run every query command; so either you need to:
a) Add it to the list of skipped command in qmp-cmd-test
b) Make it not actually error when the limit isn't enabled.
Dave
Quoted 401 lines
Quoted 399 lines
v16
- rebase on master
- drop the unused typedef syntax in [PATCH v15 6/7]
- add the Reviewed-by and Acked-by tags by the wayv15
- rebase on master
- drop the 'init_time_ms' parameter in functionvcpu_calculate_dirtyrate
- drop the 'setup' field indirtylimit_stateand calldirtylimit_processdirectly, which makes code cleaner.
- code clean indirtylimit_adjust_throttle
- fix missdirtylimit_state_unlock()indirtylimit_processanddirtylimit_query_all
- add some commentPlease review. Thanks,
Regards
Yongv14
- v13 sent by accident, resend patchset.v13
- rebase on master
- passing NULL tokvm_dirty_ring_reapin commit
"refactor per-vcpu dirty ring reaping" to keep the logic unchanged. In other word, we still try the best to reap as much PFNs as possible if dirtylimit not in service.
- move the cpu list gen id changes into a separate patch.
- release the lock before sleep during dirty page rate calculation.
- move the dirty ring size fetch logic into a separate patch.
- drop theDIRTYLIMIT_LINEAR_ADJUSTMENT_WATERMARKMACRO .
- substitute bh with function pointer when implement dirtylimit.
- merge the dirtylimit_start/stop intodirtylimit_change.
- fix "cpu-index" parameter type with "int" to keep consistency.
- fix some syntax error in documents.Please review. Thanks,
Yong
v12
- rebase on master
- add a new commmit to refactor per-vcpu dirty ring reaping, which can resolve the "vcpu miss the chances to sleep" problem
- remove thedirtylimit_threadand implemtment throttle in bottom half instead.
- let the dirty ring reaper thread keep sleeping when dirtylimit is in service
- introducecpu_list_generation_idto identifycpu_listchanging.
- keep taking thecpu_list_lockduringdirty_stat_waitto prevent vcpu plug/unplug when calculating the dirty page rate
- move the dirtylimit global initializations out ofdirtylimit_set_vcpuand do some code clean
- addDIRTYLIMIT_LINEAR_ADJUSTMENT_WATERMARKin case of oscillation when throttling
- remove the unmatched count field indirtylimit_state
- add stub to fix build on non-x86
- refactor the documentsThanks Peter and Markus for reviewing the previous versions, please review.
Thanks,
Yongv11
- rebase on master
- add a commit " refactor dirty page rate calculation" so that dirty page rate limit can reuse the calculation logic.
- handle the cpu hotplug/unplug case in the dirty page rate calculation logic.
- modify the qmp commands according to Markus's advice.
- introduce a standalone filedirtylimit.cto implement dirty page rate limit
- check if dirty limit in service bydirtylimit_statepointer instead of global variable
- introducedirtylimit_mutexto protectdirtylimit_state
- do some code clean and docsSee the commit for more detail, thanks Markus and Peter very mush for the code review and give the experienced and insightful advices, most modifications are based on these advices.
v10:
- rebase on master
- make the following modifications on patch [1/3]:
1. Make "dirtylimit-calc" thread joinable and join it after quitting.2. Add finalize function to free
dirtylimit_calc_state3. Do some code clean work
- make the following modifications on patch [2/3]:
1. Remove the original implementation of throttle according to Peter's advice.2. Introduce a negative feedback system and implement the throttle on all vcpu in one thread named "dirtylimit".
3. Simplify the algo when calculation the
throttle_us_per_full:
increase/decrease linearly when there exists a wide difference between quota and current dirty page rate, increase/decrease a fixed time slice when the difference is narrow. This makes throttle responds faster and reach the quota smoothly.4. Introduce a
unfit_cntin algo to make sure throttle really takes effect.5. Set the max sleep time 99 times more than "
ring_full_time_us".6. Make "dirtylimit" thread joinable and join it after quitting.
- make the following modifications on patch [3/3]:
1. Remove the unplug cpu handling logic.2. "query-vcpu-dirty-limit" only return dirtylimit information of vcpus that enable dirtylimit
3. Remove the "
dirtylimit_setup" function4. Trigger the dirtylimit and initialize the global state only when someone enable dirtylimit, and finalize it after the last dirtylimit be canceled.
5. Redefine the qmp command vcpu-dirty-limit/query-vcpu-dirty-limit:
enable/disable dirtylimit use a single command "vcpu-dirty-limit", to enable/disabled dirtylimit on specified vcpu only if "cpu-index" is specified, otherwise, all vcpu will be affected.6. Redefine the hmp command vcpu_dirty_limit/info
vcpu_dirty_limit- other points about the code review:
1. "merge the code of calculation dirty page rate"
I think maybe it's not suitable to touch the 'calc-dirty-rate', because 'calc-dirty-rate' will stop sync log after calculating the dirtyrate and the 'dirtylimit-cal' will not untill the last dirtylimit be canceled, if we merge theGLOBAL_DIRTY_LIMITintoGLOBAL_DIRTY_DIRTYRATE, the two are interacted with each other.2. The new implementaion of throttle algo enlightened by Peter responds faster and consume less cpu resource than the older, we make a impressed progress.
And there is a viewpoint may be discussed, it is that the new throttle logic is "passive", vcpu sleeps only after dirty ring, is full, unlike the "auto-converge" which will kick vcpu instead in a fixed slice time. If the vcpu is memory-write intensive and the ring size is large, it will produce dirty memory during the dirty ring full time and the throttle works not so good, it means the throttle depends on the dirty ring size.
I actually tested the new algo in two case:
case 1: dirty-ring-size: 4096, dirtyrate: 1170MB/s
result: minimum quota dirtyrate is 25MB/s or even less
minimum vcpu util is 6%case 2: dirty-ring-size: 65536, dirtyrate: 1170MB/s
result: minimum quota dirtyrate is 256MB/s
minimum vcpu util is 24%I post this just for discussion, i think this is not a big deal beacase if we set the dirty-ring-size to the maximum value(65536), we assume the server's bandwidth is capable of handling it.
3. I hard-code the acceptable deviation value to 25MB/s, see the macro
DIRTYLIMIT_TOLERANCE_RANGE. I'm struggling to decide whether to let it configurable4. Another point is the unplug cpu handle, current algo affects the unplugged vcpu, if we set dirty limit on it, we should fork 2 thread "dirtylimit" and "dirtylimit-calc" but do nothing, once the vcpu is hot-plugged, dirty limit works, i think the logic is ok but still there can be different advice.
- to let developers play with it easier, i post the hmp usage example:
(qemu)vcpu_dirty_limit-g on -1 500
[Please use 'infovcpu_dirty_limit' to query dirty limit for virtual CPU](qemu) info
vcpu_dirty_limit
vcpu[0], limit rate 500 (MB/s), current rate 415 (MB/s) vcpu[1], limit rate 500 (MB/s), current rate 496 (MB/s) vcpu[2], limit rate 500 (MB/s), current rate 0 (MB/s) vcpu[3], limit rate 500 (MB/s), current rate 0 (MB/s)
(qemu)vcpu_dirty_limit-g off
[Please use 'infovcpu_dirty_limit' to query dirty limit for virtual CPU](qemu) info
vcpu_dirty_limit
Dirty page limit not enabled!(qemu)
vcpu_dirty_limiton 0 300
[Please use 'infovcpu_dirty_limit' to query dirty limit for virtual CPU](qemu)
vcpu_dirty_limiton 1 500
[Please use 'infovcpu_dirty_limit' to query dirty limit for virtual CPU](qemu) info
vcpu_dirty_limit
vcpu[0], limit rate 300 (MB/s), current rate 342 (MB/s) vcpu[1], limit rate 500 (MB/s), current rate 485 (MB/s)(qemu)
vcpu_dirty_limitoff 0
[Please use 'infovcpu_dirty_limit' to query dirty limit for virtual CPU](qemu) info
vcpu_dirty_limit
vcpu[1], limit rate 500 (MB/s), current rate 528 (MB/s)(qemu)
vcpu_dirty_limitoff 1
[Please use 'infovcpu_dirty_limit' to query dirty limit for virtual CPU](qemu) info
vcpu_dirty_limit
Dirty page limit not enabled!Thanks very much for the instructive algo suggestion given by Peter, the comment and other code reviews made by Markus.
Please review, thanks!
v9:
- rebase on master
- fix the meson directory change, keep it untouched.v8:
- rebase on master
- polish the error message and remove the "unlikely" compilation syntax according to the advice given by Markus.
- keep the dirty tracking enabled during "dirtylimit-calc" lifecycle so that the overhead can be reduced according to the advice given by Peter.
- merge the "set/cancel" qmp commands into one named "vcpu-dirty-limit" and introduce qmp command "query-vcpu-dirty-limit" to query dirty limit information about virtual CPU, according to the advice given by Peter.
- check if vcpu index is valid and handle the unplug case before enabling, disabling dirty limit for virtual CPU.
- introduce hmp commands so developers can play with them easier, use "vcpu_dirty_limit" to enable dirty limit and "infovcpu_dirty_limit" to query.The patch [2/3] has not been touched so far. Any corrections and suggetions are welcome.
Please review, thanks!
v7:
- rebase on master
- polish the comments and error message according to the advices given by Markus
- introducedirtylimit_enabledfunction to pre-check if dirty page limit is enabled before canceling.v6:
- rebase on master
- fix dirtylimit setup crash found by Markus
- polish the comments according to the advice given by Markus
- adjust the qemu qmp command tag to 7.0v5:
- rebase on master
- adjust the throttle algorithm by removing the tuning inRESTRAINT_RATIOcase so that dirty page rate could reachs the quota more quickly.
- fix percentage update in throttle iteration.v4:
- rebase on master
- modify the following points according to the advice given by Markus
1. move the defination into migration.json
2. polish the comments of set-dirty-limit
3. do the syntax check and change dirty rate to dirty page rateThanks for the carefule reviews made by Markus.
Please review, thanks!
v3:
- rebase on master
- modify the following points according to the advice given by Markus
1. remove the DirtyRateQuotaVcpu and use its field as option directly
2. add comments to show details of what dirtylimit setup do
3. explain how to use dirtylimit in combination with existing qmp commands "calc-dirty-rate" and "query-dirty-rate" in documentation.Thanks for the carefule reviews made by Markus.
Please review, thanks!
Hyman
v2:
- rebase on master
- modify the following points according to the advices given by Juan
1. rename dirtyrestraint to dirtylimit
2. implement the full lifecyle function ofdirtylimit_calc, includedirtylimit_calcanddirtylimit_calc_quit
3. introduce 'quit' field indirtylimit_calc_stateto implement thedirtylimit_calc_quit
4. remove theready_condandready_mtxsince it may not be suitable
5. put the 'record_dirtypage' function code at the beggining of the file
6. remove the unnecesary return;
- other modifications has been made after code review
1. introduce 'bmap' and 'nr' field indirtylimit_stateto record the number of running thread forked by dirtylimit
2. stop the dirtyrate calculation thread if all the dirtylimit thread are stopped
3. do some renaming worksdirtyrate calulation thread -> dirtylimit-calc dirtylimit thread -> dirtylimit-{cpu_index} function name do_dirtyrestraint -> dirtylimit_check qmp command dirty-restraint -> set-drity-limit qmp command dirty-restraint-cancel -> cancel-dirty-limit header file dirtyrestraint.h -> dirtylimit.hPlease review, thanks !
thanks for the accurate and timely advices given by Juan. we really appreciate it if corrections and suggetions about this patchset are proposed.
Best Regards !
Hyman
v1:
this patchset introduce a mechanism to impose dirty restraint on vCPU, aiming to keep the vCPU running in a certain dirtyrate given by user. dirty restraint on vCPU maybe an alternative method to implement convergence logic for live migration, which could improve guest memory performance during migration compared with traditional method in theory.For the current live migration implementation, the convergence logic throttles all vCPUs of the VM, which has some side effects. -'read processes' on vCPU will be unnecessarily penalized
- throttle increase percentage step by step, which seems struggling to find the optimal throttle percentage when dirtyrate is high.
- hard to predict the remaining time of migration if the throttling percentage reachs 99%to a certain extent, the dirty restraint machnism can fix these effects by throttling at vCPU granularity during migration.
the implementation is rather straightforward, we calculate vCPU dirtyrate via the Dirty Ring mechanism periodically as the commit
0e21bf246"implement dirty-ring dirtyrate calculation" does, for vCPU that be specified to impose dirty restraint, we throttle it periodically as the auto-converge does, once after throttling, we compare the quota dirtyrate with current dirtyrate, if current dirtyrate is not under the quota, increase the throttling percentage until current dirtyrate is under the quota.this patchset is the basis of implmenting a new auto-converge method for live migration, we introduce two qmp commands for impose/cancel the dirty restraint on specified vCPU, so it also can be an independent api to supply the upper app such as libvirt, which can use it to implement the convergence logic during live migration, supplemented with the qmp 'calc-dirty-rate' command or whatever.
we post this patchset for RFC and any corrections and suggetions about the implementation, api, throttleing algorithm or whatever are very appreciated!
Please review, thanks !
Best Regards !
Hyman Huang (7):
accel/kvm/kvm-all: Refactor per-vcpu dirty ring reaping
cpus: Introducecpu_list_generation_idmigration/dirtyrate: Refactor dirty page rate calculation softmmu/dirtylimit: Implement vCPU dirtyrate calculation periodically accel/kvm/kvm-all: Introducekvm_dirty_ring_sizefunction softmmu/dirtylimit: Implement virtual CPU throttle softmmu/dirtylimit: Implement dirty page rate limitaccel/kvm/kvm-all.c | 45 +++- accel/stubs/kvm-stub.c | 5 + cpus-common.c | 8 + hmp-commands-info.hx | 13 + hmp-commands.hx | 32 +++ include/exec/cpu-common.h | 1 + include/exec/memory.h | 5 +- include/hw/core/cpu.h | 6 + include/monitor/hmp.h | 3 + include/sysemu/dirtylimit.h | 37 +++ include/sysemu/dirtyrate.h | 28 +++ include/sysemu/kvm.h | 2 + migration/dirtyrate.c | 227 +++++++++++------ migration/dirtyrate.h | 7 +- qapi/migration.json | 80 ++++++ softmmu/dirtylimit.c | 602 ++++++++++++++++++++++++++++++++++++++++++++ softmmu/meson.build | 1 + softmmu/trace-events | 7 + 18 files changed, 1010 insertions(+), 99 deletions(-) create mode 100644 include/sysemu/dirtylimit.h create mode 100644 include/sysemu/dirtyrate.h create mode 100644 softmmu/dirtylimit.cSignature · 1 line
1.8.3.1Signature · 1 line
Dr. David Alan Gilbert / [email protected] / Manchester, UK
Signature · 1 line
Dr. David Alan Gilbert / [email protected] / Manchester, UK
Quoted 25 lines · Dr. David Alan Gilbert
Quoted 4 lines · * Dr. David Alan Gilbert ([email protected]) wrote:
Quoted 1 line · * [email protected] ([email protected]) wrote:
From: Hyman Huang(黄勇) <[email protected]>
Queued via my migration/hmp/etc tree
Hi,
Unfortunately I've had to unqueue this - it breaks the
qmp-cmd-test:# starting QEMU: exec ./x86_64-softmmu/qemu-system-x86_64 -qtest unix:/tmp/qtest-142136.sock -qtest-log /dev/fd/2 -chardev socket,path=/tmp/qtest-142136.qmp,id=char0 -mon chardev=char0,mode=control -display none -nodefaults -machine none -accel qtest [I 1646239093.713627] OPENED [R +0.000190] endianness [S +0.000196] OK little {"QMP": {"version": {"qemu": {"micro": 50, "minor": 2, "major": 6}, "package": "v6.2.0-1867-g817703d65a"}, "capabilities": ["oob"]}}{"execute": "qmp_capabilities"}{"return": {}}{"execute": "query-vcpu-dirty-limit"}
{"error": {"class": "GenericError", "desc": "dirty page limit not enabled"}}** ERROR:../tests/qtest/qmp-cmd-test.c:84:test_query: assertion failed: (
qdict_haskey(resp, "return")) Bail out! ERROR:../tests/qtest/qmp-cmd-test.c:84:test_query: assertion failed: (qdict_haskey(resp, "return")) [I +0.195433] CLOSED
Aborted (core dumped)
Quoted 2 lines
qmp-cmd-test tries to run every query command; so either you need to:
a) Add it to the list of skipped command in qmp-cmd-test
query-vcpu-dirty-limit sucess only if dirty ring feature enabled. So i prefer to add this command to the list of kipped command. I'll fix it next version and run the qtests before i post the patchset.
Thinks
Yong
Quoted 405 lines
b) Make it not actually error when the limit isn't enabled.
Dave
Quoted 401 lines
Quoted 399 lines
v16
- rebase on master
- drop the unused typedef syntax in [PATCH v15 6/7]
- add the Reviewed-by and Acked-by tags by the wayv15
- rebase on master
- drop the 'init_time_ms' parameter in functionvcpu_calculate_dirtyrate
- drop the 'setup' field indirtylimit_stateand calldirtylimit_processdirectly, which makes code cleaner.
- code clean indirtylimit_adjust_throttle
- fix missdirtylimit_state_unlock()indirtylimit_processanddirtylimit_query_all
- add some commentPlease review. Thanks,
Regards
Yongv14
- v13 sent by accident, resend patchset.v13
- rebase on master
- passing NULL tokvm_dirty_ring_reapin commit
"refactor per-vcpu dirty ring reaping" to keep the logic unchanged. In other word, we still try the best to reap as much PFNs as possible if dirtylimit not in service.
- move the cpu list gen id changes into a separate patch.
- release the lock before sleep during dirty page rate calculation.
- move the dirty ring size fetch logic into a separate patch.
- drop theDIRTYLIMIT_LINEAR_ADJUSTMENT_WATERMARKMACRO .
- substitute bh with function pointer when implement dirtylimit.
- merge the dirtylimit_start/stop intodirtylimit_change.
- fix "cpu-index" parameter type with "int" to keep consistency.
- fix some syntax error in documents.Please review. Thanks,
Yong
v12
- rebase on master
- add a new commmit to refactor per-vcpu dirty ring reaping, which can resolve the "vcpu miss the chances to sleep" problem
- remove thedirtylimit_threadand implemtment throttle in bottom half instead.
- let the dirty ring reaper thread keep sleeping when dirtylimit is in service
- introducecpu_list_generation_idto identifycpu_listchanging.
- keep taking thecpu_list_lockduringdirty_stat_waitto prevent vcpu plug/unplug when calculating the dirty page rate
- move the dirtylimit global initializations out ofdirtylimit_set_vcpuand do some code clean
- addDIRTYLIMIT_LINEAR_ADJUSTMENT_WATERMARKin case of oscillation when throttling
- remove the unmatched count field indirtylimit_state
- add stub to fix build on non-x86
- refactor the documentsThanks Peter and Markus for reviewing the previous versions, please review.
Thanks,
Yongv11
- rebase on master
- add a commit " refactor dirty page rate calculation" so that dirty page rate limit can reuse the calculation logic.
- handle the cpu hotplug/unplug case in the dirty page rate calculation logic.
- modify the qmp commands according to Markus's advice.
- introduce a standalone filedirtylimit.cto implement dirty page rate limit
- check if dirty limit in service bydirtylimit_statepointer instead of global variable
- introducedirtylimit_mutexto protectdirtylimit_state
- do some code clean and docsSee the commit for more detail, thanks Markus and Peter very mush for the code review and give the experienced and insightful advices, most modifications are based on these advices.
v10:
- rebase on master
- make the following modifications on patch [1/3]:
1. Make "dirtylimit-calc" thread joinable and join it after quitting.2. Add finalize function to free
dirtylimit_calc_state3. Do some code clean work
- make the following modifications on patch [2/3]:
1. Remove the original implementation of throttle according to Peter's advice.2. Introduce a negative feedback system and implement the throttle on all vcpu in one thread named "dirtylimit".
3. Simplify the algo when calculation the
throttle_us_per_full:
increase/decrease linearly when there exists a wide difference between quota and current dirty page rate, increase/decrease a fixed time slice when the difference is narrow. This makes throttle responds faster and reach the quota smoothly.4. Introduce a
unfit_cntin algo to make sure throttle really takes effect.5. Set the max sleep time 99 times more than "
ring_full_time_us".6. Make "dirtylimit" thread joinable and join it after quitting.
- make the following modifications on patch [3/3]:
1. Remove the unplug cpu handling logic.2. "query-vcpu-dirty-limit" only return dirtylimit information of vcpus that enable dirtylimit
3. Remove the "
dirtylimit_setup" function4. Trigger the dirtylimit and initialize the global state only when someone enable dirtylimit, and finalize it after the last dirtylimit be canceled.
5. Redefine the qmp command vcpu-dirty-limit/query-vcpu-dirty-limit:
enable/disable dirtylimit use a single command "vcpu-dirty-limit", to enable/disabled dirtylimit on specified vcpu only if "cpu-index" is specified, otherwise, all vcpu will be affected.6. Redefine the hmp command vcpu_dirty_limit/info
vcpu_dirty_limit- other points about the code review:
1. "merge the code of calculation dirty page rate"
I think maybe it's not suitable to touch the 'calc-dirty-rate', because 'calc-dirty-rate' will stop sync log after calculating the dirtyrate and the 'dirtylimit-cal' will not untill the last dirtylimit be canceled, if we merge theGLOBAL_DIRTY_LIMITintoGLOBAL_DIRTY_DIRTYRATE, the two are interacted with each other.2. The new implementaion of throttle algo enlightened by Peter responds faster and consume less cpu resource than the older, we make a impressed progress.
And there is a viewpoint may be discussed, it is that the new throttle logic is "passive", vcpu sleeps only after dirty ring, is full, unlike the "auto-converge" which will kick vcpu instead in a fixed slice time. If the vcpu is memory-write intensive and the ring size is large, it will produce dirty memory during the dirty ring full time and the throttle works not so good, it means the throttle depends on the dirty ring size.
I actually tested the new algo in two case:
case 1: dirty-ring-size: 4096, dirtyrate: 1170MB/s
result: minimum quota dirtyrate is 25MB/s or even less
minimum vcpu util is 6%case 2: dirty-ring-size: 65536, dirtyrate: 1170MB/s
result: minimum quota dirtyrate is 256MB/s
minimum vcpu util is 24%I post this just for discussion, i think this is not a big deal beacase if we set the dirty-ring-size to the maximum value(65536), we assume the server's bandwidth is capable of handling it.
3. I hard-code the acceptable deviation value to 25MB/s, see the macro
DIRTYLIMIT_TOLERANCE_RANGE. I'm struggling to decide whether to let it configurable4. Another point is the unplug cpu handle, current algo affects the unplugged vcpu, if we set dirty limit on it, we should fork 2 thread "dirtylimit" and "dirtylimit-calc" but do nothing, once the vcpu is hot-plugged, dirty limit works, i think the logic is ok but still there can be different advice.
- to let developers play with it easier, i post the hmp usage example:
(qemu)vcpu_dirty_limit-g on -1 500
[Please use 'infovcpu_dirty_limit' to query dirty limit for virtual CPU](qemu) info
vcpu_dirty_limit
vcpu[0], limit rate 500 (MB/s), current rate 415 (MB/s) vcpu[1], limit rate 500 (MB/s), current rate 496 (MB/s) vcpu[2], limit rate 500 (MB/s), current rate 0 (MB/s) vcpu[3], limit rate 500 (MB/s), current rate 0 (MB/s)
(qemu)vcpu_dirty_limit-g off
[Please use 'infovcpu_dirty_limit' to query dirty limit for virtual CPU](qemu) info
vcpu_dirty_limit
Dirty page limit not enabled!(qemu)
vcpu_dirty_limiton 0 300
[Please use 'infovcpu_dirty_limit' to query dirty limit for virtual CPU](qemu)
vcpu_dirty_limiton 1 500
[Please use 'infovcpu_dirty_limit' to query dirty limit for virtual CPU](qemu) info
vcpu_dirty_limit
vcpu[0], limit rate 300 (MB/s), current rate 342 (MB/s) vcpu[1], limit rate 500 (MB/s), current rate 485 (MB/s)(qemu)
vcpu_dirty_limitoff 0
[Please use 'infovcpu_dirty_limit' to query dirty limit for virtual CPU](qemu) info
vcpu_dirty_limit
vcpu[1], limit rate 500 (MB/s), current rate 528 (MB/s)(qemu)
vcpu_dirty_limitoff 1
[Please use 'infovcpu_dirty_limit' to query dirty limit for virtual CPU](qemu) info
vcpu_dirty_limit
Dirty page limit not enabled!Thanks very much for the instructive algo suggestion given by Peter, the comment and other code reviews made by Markus.
Please review, thanks!
v9:
- rebase on master
- fix the meson directory change, keep it untouched.v8:
- rebase on master
- polish the error message and remove the "unlikely" compilation syntax according to the advice given by Markus.
- keep the dirty tracking enabled during "dirtylimit-calc" lifecycle so that the overhead can be reduced according to the advice given by Peter.
- merge the "set/cancel" qmp commands into one named "vcpu-dirty-limit" and introduce qmp command "query-vcpu-dirty-limit" to query dirty limit information about virtual CPU, according to the advice given by Peter.
- check if vcpu index is valid and handle the unplug case before enabling, disabling dirty limit for virtual CPU.
- introduce hmp commands so developers can play with them easier, use "vcpu_dirty_limit" to enable dirty limit and "infovcpu_dirty_limit" to query.The patch [2/3] has not been touched so far. Any corrections and suggetions are welcome.
Please review, thanks!
v7:
- rebase on master
- polish the comments and error message according to the advices given by Markus
- introducedirtylimit_enabledfunction to pre-check if dirty page limit is enabled before canceling.v6:
- rebase on master
- fix dirtylimit setup crash found by Markus
- polish the comments according to the advice given by Markus
- adjust the qemu qmp command tag to 7.0v5:
- rebase on master
- adjust the throttle algorithm by removing the tuning inRESTRAINT_RATIOcase so that dirty page rate could reachs the quota more quickly.
- fix percentage update in throttle iteration.v4:
- rebase on master
- modify the following points according to the advice given by Markus
1. move the defination into migration.json
2. polish the comments of set-dirty-limit
3. do the syntax check and change dirty rate to dirty page rateThanks for the carefule reviews made by Markus.
Please review, thanks!
v3:
- rebase on master
- modify the following points according to the advice given by Markus
1. remove the DirtyRateQuotaVcpu and use its field as option directly
2. add comments to show details of what dirtylimit setup do
3. explain how to use dirtylimit in combination with existing qmp commands "calc-dirty-rate" and "query-dirty-rate" in documentation.Thanks for the carefule reviews made by Markus.
Please review, thanks!
Hyman
v2:
- rebase on master
- modify the following points according to the advices given by Juan
1. rename dirtyrestraint to dirtylimit
2. implement the full lifecyle function ofdirtylimit_calc, includedirtylimit_calcanddirtylimit_calc_quit
3. introduce 'quit' field indirtylimit_calc_stateto implement thedirtylimit_calc_quit
4. remove theready_condandready_mtxsince it may not be suitable
5. put the 'record_dirtypage' function code at the beggining of the file
6. remove the unnecesary return;
- other modifications has been made after code review
1. introduce 'bmap' and 'nr' field indirtylimit_stateto record the number of running thread forked by dirtylimit
2. stop the dirtyrate calculation thread if all the dirtylimit thread are stopped
3. do some renaming worksdirtyrate calulation thread -> dirtylimit-calc dirtylimit thread -> dirtylimit-{cpu_index} function name do_dirtyrestraint -> dirtylimit_check qmp command dirty-restraint -> set-drity-limit qmp command dirty-restraint-cancel -> cancel-dirty-limit header file dirtyrestraint.h -> dirtylimit.hPlease review, thanks !
thanks for the accurate and timely advices given by Juan. we really appreciate it if corrections and suggetions about this patchset are proposed.
Best Regards !
Hyman
v1:
this patchset introduce a mechanism to impose dirty restraint on vCPU, aiming to keep the vCPU running in a certain dirtyrate given by user. dirty restraint on vCPU maybe an alternative method to implement convergence logic for live migration, which could improve guest memory performance during migration compared with traditional method in theory.For the current live migration implementation, the convergence logic throttles all vCPUs of the VM, which has some side effects. -'read processes' on vCPU will be unnecessarily penalized
- throttle increase percentage step by step, which seems struggling to find the optimal throttle percentage when dirtyrate is high.
- hard to predict the remaining time of migration if the throttling percentage reachs 99%to a certain extent, the dirty restraint machnism can fix these effects by throttling at vCPU granularity during migration.
the implementation is rather straightforward, we calculate vCPU dirtyrate via the Dirty Ring mechanism periodically as the commit
0e21bf246"implement dirty-ring dirtyrate calculation" does, for vCPU that be specified to impose dirty restraint, we throttle it periodically as the auto-converge does, once after throttling, we compare the quota dirtyrate with current dirtyrate, if current dirtyrate is not under the quota, increase the throttling percentage until current dirtyrate is under the quota.this patchset is the basis of implmenting a new auto-converge method for live migration, we introduce two qmp commands for impose/cancel the dirty restraint on specified vCPU, so it also can be an independent api to supply the upper app such as libvirt, which can use it to implement the convergence logic during live migration, supplemented with the qmp 'calc-dirty-rate' command or whatever.
we post this patchset for RFC and any corrections and suggetions about the implementation, api, throttleing algorithm or whatever are very appreciated!
Please review, thanks !
Best Regards !
Hyman Huang (7):
accel/kvm/kvm-all: Refactor per-vcpu dirty ring reaping
cpus: Introducecpu_list_generation_idmigration/dirtyrate: Refactor dirty page rate calculation softmmu/dirtylimit: Implement vCPU dirtyrate calculation periodically accel/kvm/kvm-all: Introducekvm_dirty_ring_sizefunction softmmu/dirtylimit: Implement virtual CPU throttle softmmu/dirtylimit: Implement dirty page rate limitaccel/kvm/kvm-all.c | 45 +++- accel/stubs/kvm-stub.c | 5 + cpus-common.c | 8 + hmp-commands-info.hx | 13 + hmp-commands.hx | 32 +++ include/exec/cpu-common.h | 1 + include/exec/memory.h | 5 +- include/hw/core/cpu.h | 6 + include/monitor/hmp.h | 3 + include/sysemu/dirtylimit.h | 37 +++ include/sysemu/dirtyrate.h | 28 +++ include/sysemu/kvm.h | 2 + migration/dirtyrate.c | 227 +++++++++++------ migration/dirtyrate.h | 7 +- qapi/migration.json | 80 ++++++ softmmu/dirtylimit.c | 602 ++++++++++++++++++++++++++++++++++++++++++++ softmmu/meson.build | 1 + softmmu/trace-events | 7 + 18 files changed, 1010 insertions(+), 99 deletions(-)create mode 100644
include/sysemu/dirtylimit.h
create mode 100644include/sysemu/dirtyrate.h
create mode 100644softmmu/dirtylimit.cSignature · 1 line
1.8.3.1Signature · 1 line
Dr. David Alan Gilbert / [email protected] / Manchester, UK