No significant diff b/w qemux86 and qemuarm poky build of core-image-minimal


Randy MacLeod
 

Khem was asking if anyone had seen a >30% difference in build time for
qemuarm compared to other machines. I don't when building on my system:

rmacleod@yow-rmacleod-lx3 b/c-i-m-x86 $ cat x86
real 43m29.142s
user 0m19.887s
sys 0m4.110s

rmacleod@yow-rmacleod-lx3 b/c-i-m-x86 $ cat arm
real 42m58.185s
user 0m19.546s
sys 0m3.867s

This machine is relatively isolated in that there are no other users.
The arm build was done second and I didn't pre-populate the downloads
but so that may account for the 30 second difference.
I wasn't using sstate for either build.
--
# Randy MacLeod
# Wind River Linux


Khem Raj
 

On 3/30/21 10:45 AM, Randy MacLeod wrote:
Khem  was asking if anyone had seen a >30% difference in build time for
qemuarm compared to other machines. I don't when building on my system:
rmacleod@yow-rmacleod-lx3 b/c-i-m-x86 $ cat x86
real    43m29.142s
user    0m19.887s
sys     0m4.110s
rmacleod@yow-rmacleod-lx3 b/c-i-m-x86 $ cat arm
real    42m58.185s
user    0m19.546s
sys     0m3.867s
This machine is relatively isolated in that there are no other users.
The arm build was done second and I didn't pre-populate the downloads
but so that may account for the 30 second difference.
I wasn't using sstate for either build.
Thanks for this infor Randy, really appreciated, I think I will also look at the h/w angle.


Ross Burton
 

On Tue, 30 Mar 2021 at 20:41, Khem Raj <raj.khem@gmail.com> wrote:
Thanks for this infor Randy, really appreciated, I think I will also
look at the h/w angle.
Khem: if you can reproduce this on demand with a clean build (ie
oe-init-build-env with a clean build directory, just set DL_DIR) then
enabling buildstats would give some interesting metrics.
buildstats-diff can compare two build runs and tell you what the
differences are.

Ross


Khem Raj
 

On 3/30/21 1:48 PM, Ross Burton wrote:
On Tue, 30 Mar 2021 at 20:41, Khem Raj <raj.khem@gmail.com> wrote:
Thanks for this infor Randy, really appreciated, I think I will also
look at the h/w angle.
Khem: if you can reproduce this on demand with a clean build (ie
oe-init-build-env with a clean build directory, just set DL_DIR) then
enabling buildstats would give some interesting metrics.
buildstats-diff can compare two build runs and tell you what the
differences are.
yeah this is doing world build of not only core but several other layers included a usual build takes around 10 hrs for this machine config, I am planning to

1. Collect build stats
2. buiid another qemu machine on this VM

it could very well be rotting hardware underneath the VM.

Ross


Khem Raj
 

On Tue, Mar 30, 2021 at 2:20 PM Khem Raj <raj.khem@gmail.com> wrote:



On 3/30/21 1:48 PM, Ross Burton wrote:
On Tue, 30 Mar 2021 at 20:41, Khem Raj <raj.khem@gmail.com> wrote:
Thanks for this infor Randy, really appreciated, I think I will also
look at the h/w angle.
Khem: if you can reproduce this on demand with a clean build (ie
oe-init-build-env with a clean build directory, just set DL_DIR) then
enabling buildstats would give some interesting metrics.
buildstats-diff can compare two build runs and tell you what the
differences are.
yeah this is doing world build of not only core but several other layers
included a usual build takes around 10 hrs for this machine config, I am
planning to

1. Collect build stats
2. buiid another qemu machine on this VM

it could very well be rotting hardware underneath the VM.
To close the loop, it turned out to be the case that CPU performance
on VM has deteriorated
which stress-ng showed. The VM has been rebuilt and everything seems
to be holding fine now.
Thanks for your help and chiming in.

Ross