Arm Travis jobs run much slower than equivalent jobs on x86

Environments Multi CPU Architecture
1

We observed most of Arm jobs run much slower than x86. This is an CI report, including x86 and arm jobs.
https://travis-ci.com/github/MarcoDWei/ovs/builds/152933388?utm_medium=notification&utm_source=email

Some arm jobs cost x2-3 running time than x86 jobs. Is there known resource issue on Arm Travis?
8.6 amd64 KERNEL_LIST=”5.0 4.20 4.19 4.18 4.17 4.16” 8min 3sec
8.27 arm KERNEL_LIST=”5.0 4.20 4.19 4.18 4.17 4.16” 22min 53sec
8.7 amd64 KERNEL_LIST=”4.15 4.14 4.9 4.4 3.19 3.16” 7min 51sec
8.28 arm KERNEL_LIST=”4.15 4.14 4.9 4.4 3.19 3.16” 18min 18sec

1 Like
2

What makes you think it shouldn’t take different time? It’s different hardware, in a different environment, even a different virtualization technology is used.

3

The ARM64 environment was having problems on Thursday (yesterday). AMD64, PPC64 and s390x were OK.

I think the Travis team got the issue fixed. Or at least I stopped seeing the failures on Friday (today).

Maybe you can restart a failed job and see if the issue is resolved.

4

Thanks. But I re-run the cases, running time has no change.

5

Sure, It is reasonable to have different runtimes for different platforms. But I still want to know if there is possibility of improving the time cost for Arm CI.

6

Is there known resource issue on Arm Travis?

Yes. it’s a known issue. You see some similar “ARM slow” issues on “Multi CPU Architecture” category.

2 Likes