*virtio-blk: add multiread support
v2 cover *virtio-blk: add multiread support
this series adds the long missing multiread support to virtio-blk.
some remarks:
- i introduced rd_merged and wr_merged block accounting stats to blockstats as a generic interface which can be set from any driver that will introduce multirequest merging in the future.
- the knob to disable request merging is not yet there. I would add it to the device properties also as a generic interface to have the same switch for any driver that might introduce request merging in the future. As there has been no knob in the past I would post this as a seperate series as it needs some mangling in parameter parsing which might lead to further discussions.
- the old multiwrite interface is still there and might be removed.
v1->v2:
- add overflow checking for nb_sectors [Kevin]
- do not change the name of the macro of max mergable requests. [Fam]
RFC v2->v1:
- added Erics grammar corrections to Patch 4
- Patch 4
- reworked the IF tree to a SWITCH statement to make it easier to understand in virtio_blk_handle_request
- stop merging if requests switch from read to write or vice versa. e..g, otherwise we could introduce big delays as a small read request could be followed by a lot of write requests and the read requests sits there until the queue is empty.
RFC v1->RFC v2:
- completed Patch 1 by the example in qmp-commands.hx [Eric]
- used bool for merge in Patch 4 [Max]
- fixed a few typos in the commit msg of Patch 4 [Max]
- do not start merging and directly pass req[0]->qiov in case of num_reqs == 1
- avoid allocating memory for the multireq [Kevin]
- do not import block_int.h and add appropiate iface to block-backend [Kevin]
- removed debug output and added trace event for multireq [Kevin]
- fixed alloc hint for the merge qiov [Kevin]
- currently did not split virtio_submit_multireq into rw code since the redundant code would now be much bigger part than in the original patch.
- added a merge_qiov to VirtioBlockRequest. Abusing the qiov was not possible because it is initialized externally with guest memory [Kevin]
- added a pointer to VirtioBlockRequest to create a linked list of VirtioBlockBlockRequests. This list is used to complete all requests belonging to a multireq [Kevin]
Peter Lieven (4):
block: add accounting for merged requests hw/virtio-blk: add a constant for max number of merged requests
block-backend: expose bs->bl.max_transfer_length
virtio-blk: introduce multiread
block.c | 2 + block/accounting.c | 7 ++ block/block-backend.c | 5 + block/qapi.c | 2 + hmp.c | 6 +- hw/block/dataplane/virtio-blk.c | 6 +- hw/block/virtio-blk.c | 242 ++++++++++++++++++++++++--------------- include/block/accounting.h | 3 + include/hw/virtio/virtio-blk.h | 20 +++- include/sysemu/block-backend.h | 1 + qapi/block-core.json | 9 +- qmp-commands.hx | 22 +++- trace-events | 1 + 13 files changed, 216 insertions(+), 110 deletions(-)
Signature · 1 line
1.7.9.5
PATCH v2 1/4 block: add accounting for merged requests
--- block.c | 2 ++ block/accounting.c | 7 +++++++ block/qapi.c | 2 ++ hmp.c | 6 +++++- include/block/accounting.h | 3 +++ qapi/block-core.json | 9 ++++++++- qmp-commands.hx | 22 ++++++++++++++++++---- 7 files changed, 45 insertions(+), 6 deletions(-)
block.c +2 -0
@@ -4536,6 +4536,8 @@ static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs, } } +block_acct_merge_done (&bs->stats,BLOCK_ACCT_WRITE , num_reqs - outidx -1 ); +return outidx +1 ; }
block/accounting.c +7 -0
@@ -54,3 +54,10 @@ void block_acct_highest_sector(BlockAcctStats *stats, int64_t sector_num, stats->wr_highest_sector = sector_num + nb_sectors -1 ; } } + +void block_acct_merge_done (BlockAcctStats *stats,enum BlockAcctType type, +int num_requests) +{ +assert (type <BLOCK_MAX_IOTYPE ); + stats->merged[type] += num_requests; +}
block/qapi.c +2 -0
@@ -329,6 +329,8 @@ static BlockStats *bdrv_query_stats(const BlockDriverState *bs, s->stats->wr_bytes = bs->stats.nr_bytes[BLOCK_ACCT_WRITE ]; s->stats->rd_operations = bs->stats.nr_ops[BLOCK_ACCT_READ ]; s->stats->wr_operations = bs->stats.nr_ops[BLOCK_ACCT_WRITE ]; + s->stats->rd_merged = bs->stats.merged[BLOCK_ACCT_READ ]; + s->stats->wr_merged = bs->stats.merged[BLOCK_ACCT_WRITE ]; s->stats->wr_highest_offset = bs->stats.wr_highest_sector *BDRV_SECTOR_SIZE ; s->stats->flush_operations = bs->stats.nr_ops[BLOCK_ACCT_FLUSH ];
hmp.c +5 -1
@@ -474,6 +474,8 @@ void hmp_info_blockstats(Monitor *mon, const QDict *qdict)" wr_total_time_ns=%" PRId64" rd_total_time_ns=%" PRId64" flush_total_time_ns=%" PRId64 +" rd_merged=%" PRId64 +" wr_merged=%" PRId64"\n" , stats->value->stats->rd_bytes, stats->value->stats->wr_bytes,
@@ -482,7 +484,9 @@ void hmp_info_blockstats(Monitor *mon, const QDict *qdict) stats->value->stats->flush_operations, stats->value->stats->wr_total_time_ns, stats->value->stats->rd_total_time_ns, - stats->value->stats->flush_total_time_ns); + stats->value->stats->flush_total_time_ns, + stats->value->stats->rd_merged, + stats->value->stats->wr_merged); }qapi_free_BlockStatsList (stats_list);
include/block/accounting.h +3 -0
@@ -39,6 +39,7 @@ typedef struct BlockAcctStats {uint64_t nr_bytes[BLOCK_MAX_IOTYPE ];uint64_t nr_ops[BLOCK_MAX_IOTYPE ];uint64_t total_time_ns[BLOCK_MAX_IOTYPE ]; +uint64_t merged[BLOCK_MAX_IOTYPE ];uint64_t wr_highest_sector; } BlockAcctStats;
@@ -53,5 +54,7 @@ void block_acct_start(BlockAcctStats *stats, BlockAcctCookie *cookie,void block_acct_done (BlockAcctStats *stats,BlockAcctCookie *cookie);void block_acct_highest_sector (BlockAcctStats *stats,int64_t sector_num,unsigned int nb_sectors); +void block_acct_merge_done (BlockAcctStats *stats,enum BlockAcctType type, +int num_requests);#endif
qapi/block-core.json +8 -1
@@ -407,13 +407,20 @@ # growable sparse files (like qcow2) that are used on top # of a physical device. # +# @rd_merged: Number of read requests that have been merged into another +# request (Since 2.3). +# +# @wr_merged: Number of write requests that have been merged into another +# request (Since 2.3). +# # Since: 0.14.0 ## { 'type': 'BlockDeviceStats', 'data': {'rd_bytes': 'int', 'wr_bytes': 'int', 'rd_operations': 'int', 'wr_operations': 'int', 'flush_operations': 'int', 'flush_total_time_ns': 'int', 'wr_total_time_ns': 'int', - 'rd_total_time_ns': 'int', 'wr_highest_offset': 'int' } } + 'rd_total_time_ns': 'int', 'wr_highest_offset': 'int', + 'rd_merged': 'int', 'wr_merged': 'int' } } ## # @BlockStats:
qmp-commands.hx +18 -4
@@ -2261,6 +2261,10 @@ Each json-object contain the following: - "flush_total_time_ns": total time spend on cache flushes in nano-seconds (json-int) - "wr_highest_offset": Highest offset of a sector written since the BlockDriverState has been opened (json-int) + - "rd_merged": number of read requests that have been merged into + another request (json-int) + - "wr_merged": number of write requests that have been merged into + another request (json-int) - "parent": Contains recursively the statistics of the underlying protocol (e.g. the host file for a qcow2 image). If there is no underlying protocol, this field is omitted
@@ -2284,6 +2288,8 @@ Example: "rd_total_times_ns":3465673657 "flush_total_times_ns":49653 "flush_operations":61, + "rd_merged":0, + "wr_merged":0 } }, "stats":{
@@ -2295,7 +2301,9 @@ Example: "flush_operations":51, "wr_total_times_ns":313253456 "rd_total_times_ns":3465673657 - "flush_total_times_ns":49653 + "flush_total_times_ns":49653, + "rd_merged":0, + "wr_merged":0 } }, {
@@ -2309,7 +2317,9 @@ Example: "flush_operations":0, "wr_total_times_ns":0 "rd_total_times_ns":0 - "flush_total_times_ns":0 + "flush_total_times_ns":0, + "rd_merged":0, + "wr_merged":0 } }, {
@@ -2323,7 +2333,9 @@ Example: "flush_operations":0, "wr_total_times_ns":0 "rd_total_times_ns":0 - "flush_total_times_ns":0 + "flush_total_times_ns":0, + "rd_merged":0, + "wr_merged":0 } }, {
@@ -2337,7 +2349,9 @@ Example: "flush_operations":0, "wr_total_times_ns":0 "rd_total_times_ns":0 - "flush_total_times_ns":0 + "flush_total_times_ns":0, + "rd_merged":0, + "wr_merged":0 } } ]
Signature · 1 line
1.7.9.5
PATCH v2 2/4 hw/virtio-blk: add a constant for max number of merged requests
In reply to Peter LievenAs it was not obvious (at least for me) where the 32 comes from; add a constant for it.
--- hw/block/virtio-blk.c | 2 +- include/hw/virtio/virtio-blk.h | 4 +++- 2 files changed, 4 insertions(+), 2 deletions(-)
hw/block/virtio-blk.c +1 -1
@@ -326,7 +326,7 @@ static void virtio_blk_handle_write(VirtIOBlockReq *req, MultiReqBuffer *mrb)block_acct_start (blk_get_stats (req->dev->blk), &req->acct, req->qiov.size,BLOCK_ACCT_WRITE ); -if (mrb->num_writes ==32 ) { +if (mrb->num_writes ==VIRTIO_BLK_MAX_MERGE_REQS ) {virtio_submit_multiwrite (req->dev->blk, mrb); }
include/hw/virtio/virtio-blk.h +3 -1
@@ -134,8 +134,10 @@ typedef struct VirtIOBlock {struct VirtIOBlockDataPlane *dataplane; } VirtIOBlock; +#define VIRTIO_BLK_MAX_MERGE_REQS 32 +typedef struct MultiReqBuffer { -BlockRequest blkreq[32 ]; +BlockRequest blkreq[VIRTIO_BLK_MAX_MERGE_REQS ];unsigned int num_writes; }MultiReqBuffer ;
Signature · 1 line
1.7.9.5
PATCH v2 3/4 block-backend: expose bs->bl.max_transfer_length
In reply to Peter Lieven--- block/block-backend.c | 5 +++++ include/sysemu/block-backend.h | 1 + 2 files changed, 6 insertions(+)
block/block-backend.c +5 -0
@@ -584,6 +584,11 @@ int blk_get_flags(BlockBackend *blk)return bdrv_get_flags (blk->bs); } +int blk_get_max_transfer_length (BlockBackend *blk) +{ +return blk->bs->bl.max_transfer_length; +} +void blk_set_guest_block_size (BlockBackend *blk,int align) {bdrv_set_guest_block_size (blk->bs, align);
include/sysemu/block-backend.h +1 -0
@@ -127,6 +127,7 @@ int blk_is_inserted(BlockBackend *blk);void blk_lock_medium (BlockBackend *blk,bool locked);void blk_eject (BlockBackend *blk,bool eject_flag);int blk_get_flags (BlockBackend *blk); +int blk_get_max_transfer_length (BlockBackend *blk);void blk_set_guest_block_size (BlockBackend *blk,int align);void *blk_blockalign (BlockBackend *blk,size_t size);bool blk_op_is_blocked (BlockBackend *blk,BlockOpType op,Error **errp);
Signature · 1 line
1.7.9.5
PATCH v2 4/4 virtio-blk: introduce multiread
In reply to Peter Lieventhis patch finally introduces multiread support to virtio-blk. While multiwrite support was there for a long time, read support was missing.
To achieve this the patch does several things which might need further explanation:
- the whole merge and multireq logic is moved from block.c into virtio-blk. This is move is a preparation for directly creating a coroutine out of virtio-blk.
- requests are only merged if they are strictly sequential, and no longer sorted. This simplification decreases overhead and reduces latency. It will also merge some requests which were unmergable before.
The old algorithm took up to 32 requests, sorted them and tried to merge them. The outcome was anything between 1 and 32 requests. In case of 32 requests there were 31 requests unnecessarily delayed.
On the other hand let's imagine e.g. 16 unmergeable requests followed by 32 mergable requests. The latter 32 requests would have been split into two 16 byte requests.
Last the simplified logic allows for a fast path if we have only a single request in the multirequest. In this case the request is sent as ordinary request without multireq callbacks.
As a first benchmark I installed Ubuntu 14.04.1 on a local SSD. The number of merged requests is in the same order while the write latency is obviously decreased by several percent.
cmdline:
qemu-system-x86_64 -m 1024 -smp 2 -enable-kvm -cdrom ubuntu-14.04.1-server-amd64.iso \
-drive if=virtio,file=/dev/ssd/ubuntu1404,aio=native,cache=none -monitor stdio
Before:
virtio0:
rd_bytes=151056896 wr_bytes=2683947008 rd_operations=18614 wr_operations=67979 flush_operations=15335 wr_total_time_ns=540428034217 rd_total_time_ns=11110520068 flush_total_time_ns=40673685006 rd_merged=0 wr_merged=15531
After:
virtio0:
rd_bytes=149487104 wr_bytes=2701344768 rd_operations=18148 wr_operations=68578 flush_operations=15368 wr_total_time_ns=437030089565 rd_total_time_ns=9836288815 flush_total_time_ns=40597981121 rd_merged=690 wr_merged=14615
Some first numbers of improved read performance while booting:
The Ubuntu 14.04.1 vServer from above:
virtio0:
rd_bytes=97545216 wr_bytes=119808 rd_operations=5071 wr_operations=26 flush_operations=2 wr_total_time_ns=8847669 rd_total_time_ns=13952575478 flush_total_time_ns=3075496 rd_merged=742 wr_merged=0
Windows 2012R2 (booted from iSCSI):
virtio0: rd_bytes=176559104 wr_bytes=61859840 rd_operations=7200 wr_operations=360
flush_operations=68 wr_total_time_ns=34344992718 rd_total_time_ns=134386844669 flush_total_time_ns=18115517 rd_merged=641 wr_merged=216
--- hw/block/dataplane/virtio-blk.c | 6 +- hw/block/virtio-blk.c | 242 ++++++++++++++++++++++++--------------- include/hw/virtio/virtio-blk.h | 22 ++-- trace-events | 1 + 4 files changed, 165 insertions(+), 106 deletions(-)
hw/block/dataplane/virtio-blk.c +2 -4
@@ -96,9 +96,7 @@ static void handle_notify(EventNotifier *e)event_notifier_test_and_clear (&s->host_notifier);blk_io_plug (s->conf->conf.blk);for (;;) { -MultiReqBuffer mrb = { - .num_writes =0 , - }; +MultiReqBuffer mrb = {};int ret;/* Disable guest->host notifies to avoid unnecessary vmexits */
@@ -120,7 +118,7 @@ static void handle_notify(EventNotifier *e)virtio_blk_handle_request (req, &mrb); } -virtio_submit_multiwrite (s->conf->conf.blk, &mrb); +virtio_submit_multireq (s->conf->conf.blk, &mrb);if (likely (ret == -EAGAIN )) {/* vring emptied */ /* Re-enable guest->host notifies and stop processing the vring.
hw/block/virtio-blk.c +148 -94
328 lines not on this page · raw message
include/hw/virtio/virtio-blk.h +14 -8
@@ -134,13 +134,6 @@ typedef struct VirtIOBlock {struct VirtIOBlockDataPlane *dataplane; } VirtIOBlock; -#define VIRTIO_BLK_MAX_MERGE_REQS 32 - -typedef struct MultiReqBuffer { -BlockRequest blkreq[VIRTIO_BLK_MAX_MERGE_REQS ]; -unsigned int num_writes; -}MultiReqBuffer ; - typedef struct VirtIOBlockReq { VirtIOBlock *dev; VirtQueueElement elem;
@@ -149,8 +142,21 @@ typedef struct VirtIOBlockReq {QEMUIOVector qiov;struct VirtIOBlockReq *next;BlockAcctCookie acct; +QEMUIOVector mr_qiov; +struct VirtIOBlockReq *mr_next; } VirtIOBlockReq; +#define VIRTIO_BLK_MAX_MERGE_REQS 32 + +typedef struct MultiReqBuffer { +VirtIOBlockReq *reqs[VIRTIO_BLK_MAX_MERGE_REQS ]; +unsigned int num_reqs; +bool is_write; +int niov; +int64_t sector_num; +int nb_sectors; +}MultiReqBuffer ; +VirtIOBlockReq *virtio_blk_alloc_request (VirtIOBlock *s);void virtio_blk_free_request (VirtIOBlockReq *req);
@@ -160,6 +166,6 @@ int virtio_blk_handle_scsi_req(VirtIOBlock *blk,void virtio_blk_handle_request (VirtIOBlockReq *req,MultiReqBuffer *mrb); -void virtio_submit_multiwrite (BlockBackend *blk,MultiReqBuffer *mrb); +void virtio_submit_multireq (BlockBackend *blk,MultiReqBuffer *mrb);#endif
trace-events +1 -0
@@ -116,6 +116,7 @@ virtio_blk_req_complete(void *req, int status) "req %p status %d" virtio_blk_rw_complete(void *req, int ret) "req %p ret %d" virtio_blk_handle_write(void *req, uint64_t sector, size_t nsectors) "req %p sector %"PRIu64" nsectors %zu" virtio_blk_handle_read(void *req, uint64_t sector, size_t nsectors) "req %p sector %"PRIu64" nsectors %zu" +virtio_blk_submit_multireq(void *mrb, int num_reqs, uint64_t sector, size_t nsectors, bool is_write) "mrb %p num_reqs %d sector %"PRIu64" nsectors %zu is_write %d" # hw/block/dataplane/virtio-blk.c virtio_blk_data_plane_start(void *s) "dataplane %p"
Signature · 1 line
1.7.9.5
Am 16.12.2014 um 16:21 hat Peter Lieven geschrieben:
Quoted 15 lines
this series adds the long missing multiread support to virtio-blk.
some remarks:
- i introducedrd_mergedandwr_mergedblock accounting stats to blockstats as a generic interface which can be set from any driver that will introduce multirequest merging in the future.
- the knob to disable request merging is not yet there. I would add it to the device properties also as a generic interface to have the same switch for any driver that might introduce request merging in the future. As there has been no knob in the past I would post this as a seperate series as it needs some mangling in parameter parsing which might lead to further discussions.
- the old multiwrite interface is still there and might be removed.
Quoted 3 lines
v1->v2:
- add overflow checking fornb_sectors[Kevin]
- do not change the name of the macro of max mergable requests. [Fam]
Diff to v1 looks good. Now I just need to check what it does to the performance. Did you run any benchmarks yourself?
Kevin
Quoted 16 lines · Kevin Wolf
Am 16.12.2014 um 16:21 hat Peter Lieven geschrieben:
Quoted 15 lines
this series adds the long missing multiread support to virtio-blk.
some remarks:
- i introducedrd_mergedandwr_mergedblock accounting stats to blockstats as a generic interface which can be set from any driver that will introduce multirequest merging in the future.
- the knob to disable request merging is not yet there. I would add it to the device properties also as a generic interface to have the same switch for any driver that might introduce request merging in the future. As there has been no knob in the past I would post this as a seperate series as it needs some mangling in parameter parsing which might lead to further discussions.
- the old multiwrite interface is still there and might be removed.
Quoted 5 lines
Quoted 3 lines
v1->v2:
- add overflow checking fornb_sectors[Kevin]
- do not change the name of the macro of max mergable requests. [Fam]Diff to v1 looks good. Now I just need to check what it does to the performance. Did you run any benchmarks yourself?
I ran several installs of Debian/Ubuntu, Booting of Windows and Linux systems. I looked at rd_total_time_ns and wr_total_time_ns and saw no increase. Ofter I even saw even a decrease.
{rd,wr}_total_time_ns measures the time from virtio_blk_handle_request to virtio_blk_rw_complete. So it seems to be a good indicator for the time spent with I/O.
What you could to is post it on the top of your fio testing stack and look at the throughput. Sequential Reads should be faster. The rest not worse.
Peter
Am 16.12.2014 um 17:00 hat Peter Lieven geschrieben:
Quoted 27 lines · Kevin Wolf
Quoted 16 lines
Am 16.12.2014 um 16:21 hat Peter Lieven geschrieben:
Quoted 15 lines
this series adds the long missing multiread support to virtio-blk.
some remarks:
- i introducedrd_mergedandwr_mergedblock accounting stats to blockstats as a generic interface which can be set from any driver that will introduce multirequest merging in the future.
- the knob to disable request merging is not yet there. I would add it to the device properties also as a generic interface to have the same switch for any driver that might introduce request merging in the future. As there has been no knob in the past I would post this as a seperate series as it needs some mangling in parameter parsing which might lead to further discussions.
- the old multiwrite interface is still there and might be removed.Quoted 5 lines
Quoted 3 lines
v1->v2:
- add overflow checking fornb_sectors[Kevin]
- do not change the name of the macro of max mergable requests. [Fam]Diff to v1 looks good. Now I just need to check what it does to the performance. Did you run any benchmarks yourself?
I ran several installs of Debian/Ubuntu, Booting of Windows and Linux systems. I looked at
rd_total_time_nsandwr_total_time_nsand saw no increase. Ofter I even saw even a decrease.
Quoted 7 lines
{rd,wr}_total_time_ns measures the time from
virtio_blk_handle_requesttovirtio_blk_rw_complete. So it seems to be a good indicator for the time spent with I/O.What you could to is post it on the top of your fio testing stack and look at the throughput. Sequential Reads should be faster. The rest not worse.
So I finally ran some fio benchmark on the series. The result for small sequential reads (4k) is quite noisy, but it seems to be improved a bit. Larger sequenial reads (64k) and random reads seem to be mostly unaffected.
For writes, however, I can see a degradation. Perhaps running multiple jobs in parallel means that we don't detect and merge sequential requests any more when they are interleaved with another sequential job. Or do you have an idea what else could have changed for writes?
Kevin
Am 18.12.2014 um 11:34 schrieb Kevin Wolf:
Quoted 39 lines
Am 16.12.2014 um 17:00 hat Peter Lieven geschrieben:
Quoted 26 lines · Kevin Wolf
Quoted 16 lines
Am 16.12.2014 um 16:21 hat Peter Lieven geschrieben:
Quoted 15 lines
this series adds the long missing multiread support to virtio-blk.
some remarks:
- i introducedrd_mergedandwr_mergedblock accounting stats to blockstats as a generic interface which can be set from any driver that will introduce multirequest merging in the future.
- the knob to disable request merging is not yet there. I would add it to the device properties also as a generic interface to have the same switch for any driver that might introduce request merging in the future. As there has been no knob in the past I would post this as a seperate series as it needs some mangling in parameter parsing which might lead to further discussions.
- the old multiwrite interface is still there and might be removed.Quoted 5 lines
Quoted 3 lines
v1->v2:
- add overflow checking fornb_sectors[Kevin]
- do not change the name of the macro of max mergable requests. [Fam]Diff to v1 looks good. Now I just need to check what it does to the performance. Did you run any benchmarks yourself?
I ran several installs of Debian/Ubuntu, Booting of Windows and Linux systems. I looked at
rd_total_time_nsandwr_total_time_nsand saw no increase. Ofter I even saw even a decrease.Quoted 7 lines
{rd,wr}_total_time_ns measures the time from
virtio_blk_handle_requesttovirtio_blk_rw_complete. So it seems to be a good indicator for the time spent with I/O.What you could to is post it on the top of your fio testing stack and look at the throughput. Sequential Reads should be faster. The rest not worse.
So I finally ran some fio benchmark on the series. The result for small sequential reads (4k) is quite noisy, but it seems to be improved a bit. Larger sequenial reads (64k) and random reads seem to be mostly unaffected.
Quoted 4 lines
For writes, however, I can see a degradation. Perhaps running multiple jobs in parallel means that we don't detect and merge sequential requests any more when they are interleaved with another sequential job. Or do you have an idea what else could have changed for writes?
Right, I do not sort anymore. If this is the reason increasing the 32 (what became VIRTIO_BLK_MAX_MERGE_REQS in Patch 2) should further increase bandwidth on master and you should see an improve if you run just one sequential read job comparing between master and the multiread patch.
Things I have in mind:
- What happens in terms of latency? If you look at wr_total_time_ns how does it look compared between master and the multiread patch?
- How artifical are a lot of multiple sequential writes to the same target? The sorting will cause delays WITHOUT increasing throughput for all cases where the request cannot be merged. You don't specifiy how big the degradation is. Maybe its a fair trade.
- Wouldn't this be solved by adding multiqueue support to virtio? I think we get this interleaving because several queues are piped through one channel.
Peter
Am 18.12.2014 um 11:34 schrieb Kevin Wolf:
Quoted 39 lines
Am 16.12.2014 um 17:00 hat Peter Lieven geschrieben:
Quoted 26 lines · Kevin Wolf
Quoted 16 lines
Am 16.12.2014 um 16:21 hat Peter Lieven geschrieben:
Quoted 15 lines
this series adds the long missing multiread support to virtio-blk.
some remarks:
- i introducedrd_mergedandwr_mergedblock accounting stats to blockstats as a generic interface which can be set from any driver that will introduce multirequest merging in the future.
- the knob to disable request merging is not yet there. I would add it to the device properties also as a generic interface to have the same switch for any driver that might introduce request merging in the future. As there has been no knob in the past I would post this as a seperate series as it needs some mangling in parameter parsing which might lead to further discussions.
- the old multiwrite interface is still there and might be removed.Quoted 5 lines
Quoted 3 lines
v1->v2:
- add overflow checking fornb_sectors[Kevin]
- do not change the name of the macro of max mergable requests. [Fam]Diff to v1 looks good. Now I just need to check what it does to the performance. Did you run any benchmarks yourself?
I ran several installs of Debian/Ubuntu, Booting of Windows and Linux systems. I looked at
rd_total_time_nsandwr_total_time_nsand saw no increase. Ofter I even saw even a decrease.Quoted 7 lines
{rd,wr}_total_time_ns measures the time from
virtio_blk_handle_requesttovirtio_blk_rw_complete. So it seems to be a good indicator for the time spent with I/O.What you could to is post it on the top of your fio testing stack and look at the throughput. Sequential Reads should be faster. The rest not worse.
So I finally ran some fio benchmark on the series. The result for small sequential reads (4k) is quite noisy, but it seems to be improved a bit. Larger sequenial reads (64k) and random reads seem to be mostly unaffected.
Quoted 4 lines
For writes, however, I can see a degradation. Perhaps running multiple jobs in parallel means that we don't detect and merge sequential requests any more when they are interleaved with another sequential job. Or do you have an idea what else could have changed for writes?
I tried to digged a little more into this topic and maybe found whats going on. If I a right you are using Kernel >= 3.17 in the guest for your tests?
Here are my test results of 4k sequential writes under a Linux 3.13 guest.
Master:
virtio-fio: rd_bytes=856064 wr_bytes=34359738368 rd_operations=209 wr_operations=8388608 flush_operations=0 wr_total_time_ns=980610962941 rd_total_time_ns=5656675 flush_total_time_ns=0 rd_merged=0 wr_merged=6533338
Multiread_v2:
virtio-fio: rd_bytes=856064 wr_bytes=34359738368 rd_operations=209 wr_operations=8388608 flush_operations=0 wr_total_time_ns=558830918737 rd_total_time_ns=6159151 flush_total_time_ns=0 rd_merged=0 wr_merged=6266824
As you can see the number of merged requests is in the same order, but the wr_total_time_ns is heavily improved!
What happened between Linux 3.13 and 3.17 is that Ming introduced the Multiqeue feature into the virtio-blk kernel code. The blk-mq developers intentionally set the number of hw_queues to 4 in Kernel 3.13 for virtio-blk. With the introduction of the Multiqeue feature in 3.17 the number of hw_queues is set the the number of virtqueues. If multique is unsupported the number is 1 and thus the number of hw_queues is also 1. So all the requests from all fio processes go into the same hw_queue and this seems to break the performance. The requests are heavily interleaved this way and as I only merge strictly sequential requests the old implementation which sorts requests wins. But it uses twice the computation time for this.
I think this needs to be fixed in the virtio-blk kernel code and if we introduce multiqueue for virtio-blk into qemu, we should set the number of virtqueues to at least 4. Maybe the number of cpus would also be a good choice?!
Peter