Skip to contentSkip to conversation

bitbake.conf: Set ZSTD_THREADS to half of cpu number

PATCH In review Robert Yang 5 participants 12 messages Nov 9, 2021 – Nov 11, 2021

lore
Share & tools
lore
#0 lore raw

cover bitbake.conf: Set ZSTD_THREADS to half of cpu number

The following changes since commit afad2f09ebbfe395f8fafce8218e26056479fe39:

meson: improve SDK's wrapper to make Wraps work (2021-11-08 22:01:53 +0000)

are available in the Git repository at:

git://git.openembedded.org/openembedded-core-contrib rbt/zstd
http://cgit.openembedded.org/openembedded-core-contrib/log/?h=rbt/zstd

Robert Yang (1):
bitbake.conf: Set ZSTD_THREADS to half of cpu number

 meta/conf/bitbake.conf | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
Signature · 1 line
2.17.1
#1 lore raw

PATCH 1/1 bitbake.conf: Set ZSTD_THREADS to half of cpu number

The original value is very easy to cause do_packge error when cpu number is larger, for example, 128 cores and 512G mem:

error: create archive failed: cpio: write failed - Cannot allocate memory"

Set the ZSTD_THREADS to half of the CPU number can avoid the error in my testing.

---
 meta/conf/bitbake.conf | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
meta/conf/bitbake.conf +1 -1
index 71c1e52ad6..46ebf5113f 100644
@@ -833,7 +833,7 @@ XZ_DEFAULTS ?= "--memlimit=${XZ_MEMLIMIT} --threads=${XZ_THREADS}"
 XZ_DEFAULTS[vardepsexclude] += "XZ_MEMLIMIT XZ_THREADS"
 
 # Default parallelism for zstd
-ZSTD_THREADS ?= "${@oe.utils.cpu_count(at_least=2)}"
+ZSTD_THREADS ?= "${@int(oe.utils.cpu_count(at_least=4)/2)}"
 ZSTD_THREADS[vardepvalue] = "1"
 
 # Limit the number of threads that OpenMP libraries will use. Otherwise they
Signature · 1 line
2.17.1
#2 lore raw

But does this mean we need to globally restrict this and make things twice slower for everyone? I'm not sure.

Alex

Quoted 43 lines · Robert Yang

The original value is very easy to cause do_packge error when cpu number is larger, for example, 128 cores and 512G mem:

error: create archive failed: cpio: write failed - Cannot allocate memory"

Set the ZSTD_THREADS to half of the CPU number can avoid the error in my testing.

---
 meta/conf/bitbake.conf | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
meta/conf/bitbake.conf +1 -2
index 71c1e52ad6..46ebf5113f 100644
@@ -833,7 +833,7 @@ XZ_DEFAULTS ?= "--memlimit=${XZ_MEMLIMIT}
--threads=${XZ_THREADS}"
 XZ_DEFAULTS[vardepsexclude] += "XZ_MEMLIMIT XZ_THREADS"
 
 # Default parallelism for zstd
-ZSTD_THREADS ?= "${@oe.utils.cpu_count(at_least=2)}"
+ZSTD_THREADS ?= "${@int(oe.utils.cpu_count(at_least=4)/2)}"
 ZSTD_THREADS[vardepvalue] = "1"
 
 # Limit the number of threads that OpenMP libraries will use. Otherwise

they
--
2.17.1

-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group. View/Reply Online (#157998):
https://lists.openembedded.org/g/openembedded-core/message/157998
Mute This Topic: https://lists.openembedded.org/mt/86926962/1686489
Group Owner: [email protected]
Unsubscribe: https://lists.openembedded.org/g/openembedded-core/unsub [ [email protected]]
-=-=-=-=-=-=-=-=-=-=-=-

#3 lore raw

Hi Alexander,

Quoted 1 line · Alexander Kanavin

But does this mean we need to globally restrict this and make things twice slower for everyone? I'm not sure.

Please see my reply to Konrad, it doesn't cause any slower, but faster in my testing.

BTW, the XZ_MEMLIMIT is 50% for command xz, but there isn't a similar option for zstd, so I have to limit ZSTD_THREADS.

// Robert

Quoted 52 lines

Alex

On Tue, 9 Nov 2021 at 09:12, Robert Yang <[email protected] <mailto:[email protected]>> wrote:

    The original value is very easy to cause do_packge error when cpu number is
    larger, for example, 128 cores and 512G mem:

    error: create archive failed: cpio: write failed - Cannot allocate memory"

    Set the ZSTD_THREADS to half of the CPU number can avoid the error in my
    testing.

    Signed-off-by: Robert Yang <[email protected]
    <mailto:[email protected]>>
    ---
      meta/conf/bitbake.conf | 2 +-
      1 file changed, 1 insertion(+), 1 deletion(-)

    diff --git a/meta/conf/bitbake.conf b/meta/conf/bitbake.conf
    index 71c1e52ad6..46ebf5113f 100644
    --- a/meta/conf/bitbake.conf
    +++ b/meta/conf/bitbake.conf
    @@ -833,7 +833,7 @@ XZ_DEFAULTS ?= "--memlimit=${XZ_MEMLIMIT}
    --threads=${XZ_THREADS}"
      XZ_DEFAULTS[vardepsexclude] += "XZ_MEMLIMIT XZ_THREADS"

      # Default parallelism for zstd
    -ZSTD_THREADS ?= "${@oe.utils.cpu_count(at_least=2)}"
    +ZSTD_THREADS ?= "${@int(oe.utils.cpu_count(at_least=4)/2)}"
      ZSTD_THREADS[vardepvalue] = "1"

      # Limit the number of threads that OpenMP libraries will use. Otherwise they
    -- 
    2.17.1


    -=-=-=-=-=-=-=-=-=-=-=-
    Links: You receive all messages sent to this group.
    View/Reply Online (#157998):
    https://lists.openembedded.org/g/openembedded-core/message/157998
    <https://urldefense.com/v3/__https://lists.openembedded.org/g/openembedded-core/message/157998__;!!AjveYdw8EvQ!JYlw1H6vGmlFiU7pGKOBAwXZBD-wGB5MiKIahgkPJl3ojXBUC4Rkt3vdHS9iN8a9RUvg$>
    Mute This Topic: https://lists.openembedded.org/mt/86926962/1686489
    <https://urldefense.com/v3/__https://lists.openembedded.org/mt/86926962/1686489__;!!AjveYdw8EvQ!JYlw1H6vGmlFiU7pGKOBAwXZBD-wGB5MiKIahgkPJl3ojXBUC4Rkt3vdHS9iN7BIPv2n$>
    Group Owner: [email protected]
    <mailto:openembedded-core%[email protected]>
    Unsubscribe: https://lists.openembedded.org/g/openembedded-core/unsub
    <https://urldefense.com/v3/__https://lists.openembedded.org/g/openembedded-core/unsub__;!!AjveYdw8EvQ!JYlw1H6vGmlFiU7pGKOBAwXZBD-wGB5MiKIahgkPJl3ojXBUC4Rkt3vdHS9iN_MzaRNb$>
    [[email protected] <mailto:[email protected]>]
    -=-=-=-=-=-=-=-=-=-=-=-
#4 lore raw
In reply to Robert Yang
Quoted 18 lines · Robert Yang

The original value is very easy to cause do_packge error when cpu number is larger, for example, 128 cores and 512G mem:

error: create archive failed: cpio: write failed - Cannot allocate memory"

Set the ZSTD_THREADS to half of the CPU number can avoid the error in my testing.

---
  meta/conf/bitbake.conf | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
meta/conf/bitbake.conf +0 -0
index 71c1e52ad6..46ebf5113f 100644

meta/conf/bitbake.conf XZ_DEFAULTS ?= "--memlimit=${XZ_MEMLIMIT} --threads=${XZ_THREADS}" · line 836

Quoted 5 lines
meta/conf/bitbake.conf +1 -1
@@ -833,7 +833,7 @@ XZ_DEFAULTS ?= "--memlimit=${XZ_MEMLIMIT} --threads=${XZ_THREADS}"
  XZ_DEFAULTS[vardepsexclude] += "XZ_MEMLIMIT XZ_THREADS"
 
  # Default parallelism for zstd
-ZSTD_THREADS ?= "${@oe.utils.cpu_count(at_least=2)}"
+ZSTD_THREADS ?= "${@int(oe.utils.cpu_count(at_least=4)/2)}"

Then why not just limit it for the large setups you are referring to in the example, for instance like

ZSTD_THREADS ?= "${@min(int(oe.utils.cpu_count(at_least=4)), <insert a sane value of your choice>)}"

BTW this can be also done in your local.conf - as Alex already said, there is simply no reason to make it slower for everyone, while also half the threads of a 128 core machine could be a trouble some setup.

Last time I had this issue (with XZ back then) I used "${@min(int(oe.utils.cpu_count(at_least=4)), 20}"

Quoted 14 lines
meta/conf/bitbake.conf +0 -1
  ZSTD_THREADS[vardepvalue] = "1"
 
  # Limit the number of threads that OpenMP libraries will use. Otherwise they
 
 
 
-=-=-=-=-=-=-=-=-=-=-=-

Links: You receive all messages sent to this group.
View/Reply Online (#157998): https://lists.openembedded.org/g/openembedded-core/message/157998
Mute This Topic: https://lists.openembedded.org/mt/86926962/3647476
Group Owner: [email protected]
Unsubscribe: https://lists.openembedded.org/g/openembedded-core/unsub [[email protected]]
-=-=-=-=-=-=-=-=-=-=-=-

#5 lore raw
Quoted 29 lines · Konrad Weihmann
Quoted 21 lines · Robert Yang

The original value is very easy to cause do_packge error when cpu number is larger, for example, 128 cores and 512G mem:

error: create archive failed: cpio: write failed - Cannot allocate memory"

Set the ZSTD_THREADS to half of the CPU number can avoid the error in my testing.

---
  meta/conf/bitbake.conf | 2 +-

  1 file changed, 1 insertion(+), 1 deletion(-)
meta/conf/bitbake.conf +0 -1
index 71c1e52ad6..46ebf5113f 100644
@@ -833,7 +833,7 @@ XZ_DEFAULTS ?= "--memlimit=${XZ_MEMLIMIT} 
--threads=${XZ_THREADS}"

  XZ_DEFAULTS[vardepsexclude] += "XZ_MEMLIMIT XZ_THREADS" # Default parallelism for zstd

meta/conf/bitbake.conf XZ_DEFAULTS ?= "--memlimit=${XZ_MEMLIMIT} --threads=${XZ_THREADS}" · line 836

Quoted 2 lines
meta/conf/bitbake.conf +1 -1
-ZSTD_THREADS ?= "${@oe.utils.cpu_count(at_least=2)}"
+ZSTD_THREADS ?= "${@int(oe.utils.cpu_count(at_least=4)/2)}"

Then why not just limit it for the large setups you are referring to in the example, for instance like

Quoted 1 line

ZSTD_THREADS ?= "${@min(int(oe.utils.cpu_count(at_least=4)), <insert a sane value of your choice>)}"

This sounds like a good choice.

Quoted 2 lines

BTW this can be also done in your local.conf - as Alex already said, there is

Set it in my local.conf can make it work, but it isn't good for oe-core's OOBE since the issue still exists.

Quoted 1 line

simply no reason to make it slower for everyone, while also half the threads of

The whole build contains a lot of tasks, for lower cpu cores, the total resources are fixed, so when zstd uses less mem, other tasks can use more.

Limit zstd's thread doesn't make the build slower, but faster in my testing, it would be great if you can help test it. The simple testing commands are:

# Make everything ready:

$ bitbake linux-yocto

# Before the patch, set ZSTD_THREADS to 128 and run 6 times:

$ for i in `seq 0 5`; do time bitbake linux-yocto -cpackage_write_rpm -f > before_$i.log; done

Note, the time will be printed on the screen, not the before.log, the before.log is for strip the logs and then we can read the time easier.

What I get is:

real    2m12.079s
real    2m0.177s
real    1m52.426s
real    2m3.396s
real    2m16.018s
real    1m58.595s

Drop the first build time since it *may* contain parsing time, and for the last five builds:

Total: 609 seconds
Average: 609 / 5.0 = 121.8

# After the patch, set ZSTD_THREADS to 64, and run 6 times:

$ for i in `seq 0 5`; do time bitbake linux-yocto -cpackage_write_rpm -f > after_$i.log; done

What I get is:

real    1m50.017s
real    1m50.400s
real    1m53.174s
real    2m4.817s
real    1m53.476s
real    1m56.794s

Drop the first build time since it *may* contain parsing time, and for the last five builds:

Total: 576 seconds
Average: 576 / 5.0 = 115.2

So the smaller number is faster than the larger one.

// Robert

Quoted 17 lines

a 128 core machine could be a trouble some setup.

Last time I had this issue (with XZ back then) I used "${@min(int(oe.utils.cpu_count(at_least=4)), 20}"

Quoted 13 lines

  ZSTD_THREADS[vardepvalue] = "1"
  # Limit the number of threads that OpenMP libraries will use. Otherwise they

-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#157998): https://lists.openembedded.org/g/openembedded-core/message/157998
Mute This Topic: https://lists.openembedded.org/mt/86926962/3647476
Group Owner: [email protected]
Unsubscribe: https://lists.openembedded.org/g/openembedded-core/unsub [[email protected]]
-=-=-=-=-=-=-=-=-=-=-=-

#6 lore raw

Maybe, but once we start doing tweaks like this, where do they stop? Hey, I'd like to limit llvm and webkit threads too, they eat disproportionally large amounts of RAM compared to everything else! I'd rather err on the side of simple, striaghtforward, universal defaults, and having better documentation for tweaking system resource consumption. Each build system is different.

Alex

Quoted 135 lines · Robert Yang
Quoted 4 lines · Konrad Weihmann
Quoted 1 line · Robert Yang

The original value is very easy to cause do_packge error when cpu

number is

Quoted 3 lines
Quoted 3 lines

larger, for example, 128 cores and 512G mem:

error: create archive failed: cpio: write failed - Cannot allocate

memory"

Quoted 9 lines
Quoted 9 lines

Set the ZSTD_THREADS to half of the CPU number can avoid the error in my testing.

---
  meta/conf/bitbake.conf | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
Quoted 12 lines

meta/conf/bitbake.conf XZ_DEFAULTS ?= "--memlimit=${XZ_MEMLIMIT} · line 835

Quoted 10 lines
meta/conf/bitbake.conf +1 -2
index 71c1e52ad6..46ebf5113f 100644
@@ -833,7 +833,7 @@ XZ_DEFAULTS ?= "--memlimit=${XZ_MEMLIMIT}
--threads=${XZ_THREADS}"
  XZ_DEFAULTS[vardepsexclude] += "XZ_MEMLIMIT XZ_THREADS"
  # Default parallelism for zstd
-ZSTD_THREADS ?= "${@oe.utils.cpu_count(at_least=2)}"
+ZSTD_THREADS ?= "${@int(oe.utils.cpu_count(at_least=4)/2)}"

Then why not just limit it for the large setups you are referring to in

the

Quoted 3 lines

example, for instance like

ZSTD_THREADS ?= "${@min(int(oe.utils.cpu_count(at_least=4)), <insert a

sane

Quoted 1 line

value of your choice>)}"

This sounds like a good choice.

Quoted 2 lines

BTW this can be also done in your local.conf - as Alex already said,

there is

Set it in my local.conf can make it work, but it isn't good for oe-core's OOBE
since the issue still exists.

Quoted 1 line

simply no reason to make it slower for everyone, while also half the

threads of

The whole build contains a lot of tasks, for lower cpu cores, the total resources are fixed, so when zstd uses less mem, other tasks can use more.

Limit zstd's thread doesn't make the build slower, but faster in my testing,
it would be great if you can help test it. The simple testing commands are:

# Make everything ready:

$ bitbake linux-yocto

# Before the patch, set ZSTD_THREADS to 128 and run 6 times:

$ for i in `seq 0 5`; do time bitbake linux-yocto -cpackage_write_rpm -f >
before_$i.log; done

Note, the time will be printed on the screen, not the before.log, the before.log
is for strip the logs and then we can read the time easier.

What I get is:

real    2m12.079s
real    2m0.177s
real    1m52.426s
real    2m3.396s
real    2m16.018s
real    1m58.595s

Drop the first build time since it *may* contain parsing time, and for the last
five builds:

Total: 609 seconds
Average: 609 / 5.0 = 121.8

# After the patch, set ZSTD_THREADS to 64, and run 6 times:

$ for i in `seq 0 5`; do time bitbake linux-yocto -cpackage_write_rpm -f >
after_$i.log; done

What I get is:

real    1m50.017s
real    1m50.400s
real    1m53.174s
real    2m4.817s
real    1m53.476s
real    1m56.794s

Drop the first build time since it *may* contain parsing time, and for the last
five builds:

Total: 576 seconds
Average: 576 / 5.0 = 115.2

So the smaller number is faster than the larger one.

// Robert

Quoted 7 lines

a 128 core machine could be a trouble some setup.

Last time I had this issue (with XZ back then) I used "${@min(int(oe.utils.cpu_count(at_least=4)), 20}"

Quoted 2 lines
meta/conf/bitbake.conf +0 -0
  ZSTD_THREADS[vardepvalue] = "1"
  # Limit the number of threads that OpenMP libraries will use.

Otherwise they

Quoted 5 lines
Quoted 5 lines

-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group. View/Reply Online (#158004):
https://lists.openembedded.org/g/openembedded-core/message/158004
Mute This Topic: https://lists.openembedded.org/mt/86926962/1686489
Group Owner: [email protected]
Unsubscribe: https://lists.openembedded.org/g/openembedded-core/unsub [ [email protected]]
-=-=-=-=-=-=-=-=-=-=-=-

#7 lore raw

Hi Alexander,

Quoted 1 line · Alexander Kanavin

Maybe, but once we start doing tweaks like this, where do they stop? Hey, I'd

I think that it's not only a performance tweak, but this should be considered as a bug since it would be failed to build on powerful machine, which isn't good for oe-core's user experience.

// Robert

Quoted 136 lines

like to limit llvm and webkit threads too, they eat disproportionally large amounts of RAM compared to everything else! I'd rather err on the side of simple, striaghtforward, universal defaults, and having better documentation for tweaking system resource consumption. Each build system is different.

Alex

On Tue, 9 Nov 2021 at 10:40, Robert Yang <[email protected] <mailto:[email protected]>> wrote:

    On 11/9/21 4:41 PM, Konrad Weihmann wrote:
     >
     >
     > On 09.11.21 09:48, Robert Yang wrote:
     >> The original value is very easy to cause do_packge error when cpu number is
     >> larger, for example, 128 cores and 512G mem:
     >>
     >> error: create archive failed: cpio: write failed - Cannot allocate memory"
     >>
     >> Set the ZSTD_THREADS to half of the CPU number can avoid the error in my
     >> testing.
     >>
     >> Signed-off-by: Robert Yang <[email protected]
    <mailto:[email protected]>>
     >> ---
     >>   meta/conf/bitbake.conf | 2 +-
     >>   1 file changed, 1 insertion(+), 1 deletion(-)
     >>
     >> diff --git a/meta/conf/bitbake.conf b/meta/conf/bitbake.conf
     >> index 71c1e52ad6..46ebf5113f 100644
     >> --- a/meta/conf/bitbake.conf
     >> +++ b/meta/conf/bitbake.conf
     >> @@ -833,7 +833,7 @@ XZ_DEFAULTS ?= "--memlimit=${XZ_MEMLIMIT}
     >> --threads=${XZ_THREADS}"
     >>   XZ_DEFAULTS[vardepsexclude] += "XZ_MEMLIMIT XZ_THREADS"
     >>   # Default parallelism for zstd
     >> -ZSTD_THREADS ?= "${@oe.utils.cpu_count(at_least=2)}"
     >> +ZSTD_THREADS ?= "${@int(oe.utils.cpu_count(at_least=4)/2)}"
     >
     > Then why not just limit it for the large setups you are referring to in the
     > example, for instance like
     >
     > ZSTD_THREADS ?= "${@min(int(oe.utils.cpu_count(at_least=4)), <insert a sane
     > value of your choice>)}"

    This sounds like a good choice.

     >
     > BTW this can be also done in your local.conf - as Alex already said,
    there is

    Set it in my local.conf can make it work, but it isn't good for oe-core's OOBE
    since the issue still exists.

     > simply no reason to make it slower for everyone, while also half the
    threads of

    The whole build contains a lot of tasks, for lower cpu cores, the total
    resources are fixed, so when zstd uses less mem, other tasks can use more.

    Limit zstd's thread doesn't make the build slower, but faster in my testing,
    it would be great if you can help test it. The simple testing commands are:

    # Make everything ready:
    $ bitbake linux-yocto

    # Before the patch, set ZSTD_THREADS to 128 and run 6 times:
    $ for i in `seq 0 5`; do time bitbake linux-yocto -cpackage_write_rpm -f >
    before_$i.log; done

    Note, the time will be printed on the screen, not the before.log, the before.log
    is for strip the logs and then we can read the time easier.

    What I get is:
    real    2m12.079s
    real    2m0.177s
    real    1m52.426s
    real    2m3.396s
    real    2m16.018s
    real    1m58.595s

    Drop the first build time since it *may* contain parsing time, and for the last
    five builds:

    Total: 609 seconds
    Average: 609 / 5.0 = 121.8


    # After the patch, set ZSTD_THREADS to 64, and run 6 times:
    $ for i in `seq 0 5`; do time bitbake linux-yocto -cpackage_write_rpm -f >
    after_$i.log; done

    What I get is:
    real    1m50.017s
    real    1m50.400s
    real    1m53.174s
    real    2m4.817s
    real    1m53.476s
    real    1m56.794s

    Drop the first build time since it *may* contain parsing time, and for the last
    five builds:

    Total: 576 seconds
    Average: 576 / 5.0 = 115.2

    So the smaller number is faster than the larger one.

    // Robert

     > a 128 core machine could be a trouble some setup.
     >
     > Last time I had this issue (with XZ back then) I used
     > "${@min(int(oe.utils.cpu_count(at_least=4)), 20}"
     >
     >>   ZSTD_THREADS[vardepvalue] = "1"
     >>   # Limit the number of threads that OpenMP libraries will use.
    Otherwise they
     >>
     >>
     >>
     >>
     >>

    -=-=-=-=-=-=-=-=-=-=-=-
    Links: You receive all messages sent to this group.
    View/Reply Online (#158004):
    https://lists.openembedded.org/g/openembedded-core/message/158004
    <https://urldefense.com/v3/__https://lists.openembedded.org/g/openembedded-core/message/158004__;!!AjveYdw8EvQ!Pp09fG5njytGigcrK1Pci9c4pmhLI2WooNDvgaDRdaq-cC5BmErvgdfHBe3ivNZBDAzT$>
    Mute This Topic: https://lists.openembedded.org/mt/86926962/1686489
    <https://urldefense.com/v3/__https://lists.openembedded.org/mt/86926962/1686489__;!!AjveYdw8EvQ!Pp09fG5njytGigcrK1Pci9c4pmhLI2WooNDvgaDRdaq-cC5BmErvgdfHBe3ivG7_2fs8$>
    Group Owner: [email protected]
    <mailto:openembedded-core%[email protected]>
    Unsubscribe: https://lists.openembedded.org/g/openembedded-core/unsub
    <https://urldefense.com/v3/__https://lists.openembedded.org/g/openembedded-core/unsub__;!!AjveYdw8EvQ!Pp09fG5njytGigcrK1Pci9c4pmhLI2WooNDvgaDRdaq-cC5BmErvgdfHBe3ivAnK9lkk$>
    [[email protected] <mailto:[email protected]>]
    -=-=-=-=-=-=-=-=-=-=-=-
#8 lore raw
Quoted 9 lines · Robert Yang
Quoted 1 line

Maybe, but once we start doing tweaks like this, where do they stop?

Hey, I'd

I think that it's not only a performance tweak, but this should be considered as
a bug since it would be failed to build on powerful machine, which isn't good
for oe-core's user experience.

I managed to bring down a powerful machine just yesterday by building webkit and llvm together at the same time. They together ate all the RAM, and someone in the office had to go and reset the box.
Do I see it as a bug? Absolutely not: it's in fact my fault for not ensuring system resource consumption doesn't overwhelm the machine. This is the same: if you do not have RAM to match the cores on your specific rig, add RAM or limit the threads.

Alex

#9 lore raw
Quoted 13 lines · Alexander Kanavin
Quoted 10 lines · Robert Yang
Quoted 1 line

Maybe, but once we start doing tweaks like this, where do they stop? Hey,

I'd

I think that it's not only a performance tweak, but this should be considered
as
a bug since it would be failed to build on powerful machine, which isn't good
for oe-core's user experience.

Quoted 7 lines

I managed to bring down a powerful machine just yesterday by building webkit and llvm together at the same time. They together ate all the RAM, and someone in the office had to go and reset the box.
Do I see it as a bug? Absolutely not: it's in fact my fault for not ensuring system resource consumption doesn't overwhelm the machine. This is the same:
if you do not have RAM to match the cores on your specific rig, add RAM or limit the threads.

I can see both sides of this. If we had a pool for all the threads and the pieces of the system all used the same pool, it would work well as implemented. Sadly the bitbake threads, the make processes and now the compression threads are all separate.

We do see the autobuilder break due to load issues too and on that we have scaled back some of the defaults but not ZSTD as yet that I recall.

As such I think limiting the upper value on number of threads may be a good compromise, both for compression and maybe for parallel make too.

Cheers,

Richard

#10 lore raw

Hi RP,

Quoted 27 lines · Richard Purdie
Quoted 13 lines · Alexander Kanavin
Quoted 10 lines · Robert Yang
Quoted 1 line

Maybe, but once we start doing tweaks like this, where do they stop? Hey,

I'd

I think that it's not only a performance tweak, but this should be considered
as
a bug since it would be failed to build on powerful machine, which isn't good
for oe-core's user experience.

Quoted 7 lines

I managed to bring down a powerful machine just yesterday by building webkit and llvm together at the same time. They together ate all the RAM, and someone in the office had to go and reset the box.
Do I see it as a bug? Absolutely not: it's in fact my fault for not ensuring system resource consumption doesn't overwhelm the machine. This is the same:
if you do not have RAM to match the cores on your specific rig, add RAM or limit the threads.

I can see both sides of this. If we had a pool for all the threads and the pieces of the system all used the same pool, it would work well as implemented. Sadly the bitbake threads, the make processes and now the compression threads are all separate.

Quoted 5 lines

We do see the autobuilder break due to load issues too and on that we have scaled back some of the defaults but not ZSTD as yet that I recall.

As such I think limiting the upper value on number of threads may be a good compromise, both for compression and maybe for parallel make too.

What should we do next, please? Limit bb thread + parallel make + zstd thread? Or just limit zstd's thread since other people may concern about the performance.

// Robert

Quoted 5 lines

Cheers,

Richard

#11 lore raw
In reply to Robert Yang

Robert Yang <[email protected]> escreveu no dia terça, 9/11/2021 à(s) 08:12:

Quoted 18 lines

The original value is very easy to cause do_packge error when cpu number is larger, for example, 128 cores and 512G mem:

error: create archive failed: cpio: write failed - Cannot allocate memory"

Set the ZSTD_THREADS to half of the CPU number can avoid the error in my testing.

---
 meta/conf/bitbake.conf | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
meta/conf/bitbake.conf +0 -0
index 71c1e52ad6..46ebf5113f 100644

meta/conf/bitbake.conf XZ_DEFAULTS ?= "--memlimit=${XZ_MEMLIMIT} · line 837

Quoted 7 lines
meta/conf/bitbake.conf +1 -2
@@ -833,7 +833,7 @@ XZ_DEFAULTS ?= "--memlimit=${XZ_MEMLIMIT}
--threads=${XZ_THREADS}"
 XZ_DEFAULTS[vardepsexclude] += "XZ_MEMLIMIT XZ_THREADS"
 
 # Default parallelism for zstd
-ZSTD_THREADS ?= "${@oe.utils.cpu_count(at_least=2)}"
+ZSTD_THREADS ?= "${@int(oe.utils.cpu_count(at_least=4)/2)}"
 

This will slow down all the systems with lower cpu cores.

Quoted 19 lines
meta/conf/bitbake.conf +0 -0
 ZSTD_THREADS[vardepvalue] = "1"
 
 # Limit the number of threads that OpenMP libraries will use. Otherwise

they
--
2.17.1

-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group. View/Reply Online (#157998):
https://lists.openembedded.org/g/openembedded-core/message/157998
Mute This Topic: https://lists.openembedded.org/mt/86926962/5052612
Group Owner: [email protected]
Unsubscribe: https://lists.openembedded.org/g/openembedded-core/unsub [ [email protected]]
-=-=-=-=-=-=-=-=-=-=-=-

Signature · 3 lines
Best regards,

José Quaresma