Skip to content

Pass nodelay flag in ucp perftest #10894

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
wants to merge 5 commits into from
Closed

Pass nodelay flag in ucp perftest #10894

Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
db52bec
UCP/PERF: Pass no delay flag in perftest.
ofirfarjun7 Sep 17, 2025
f502955
UCP/PERF: Format.
ofirfarjun7 Sep 17, 2025
dd8f962
UCP/PERF: Pass no delay flag in perftest.
ofirfarjun7 Sep 17, 2025
bb0da5e
UCP/PERF: CR comments.
ofirfarjun7 Sep 17, 2025
cd4ccdb
UCP/PERF: Revert format.
ofirfarjun7 Sep 18, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 4 additions & 2 deletions src/tools/perf/cuda/ucp_cuda_kernel.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -98,7 +98,8 @@ ucp_perf_cuda_put_multi_bw_kernel(ucx_perf_cuda_context &ctx,
*sn = idx + 1;
ucp_device_request_t &req = request_mgr.get_request();
status = ucp_device_put_single<level>(mem_list, mem_list_index, address,
remote_address, length, 0, &req);
remote_address, length,
UCP_DEVICE_FLAG_NODELAY, &req);
if (status != UCS_OK) {
break;
}
Expand Down Expand Up @@ -139,7 +140,8 @@ ucp_perf_cuda_put_single(ucp_device_mem_list_handle_h mem_list,
ucs_status_t status;

status = ucp_device_put_single<level>(mem_list, mem_list_index, address,
remote_address, length, 0, &req);
remote_address, length,
UCP_DEVICE_FLAG_NODELAY, &req);
Copy link
Contributor

@yosefe yosefe Sep 17, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

how did it pass CI?

Copy link
Contributor

@iyastreb iyastreb Sep 18, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Hmm, it works fine on my side without this flag

Copy link
Contributor

@iyastreb iyastreb Sep 18, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can we keep only this change and exclude formatting changes?

Copy link
Contributor Author

@ofirfarjun7 ofirfarjun7 Sep 18, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Did it worked for you with two nodes? With single node & cudaipc it will not go throw NIC so it will work without DB. (Maybe this is why CI passed?)

I can just add the flag and revert the rest. We need it to measure perf so I don't want to wait for PR with many changes that require review

Copy link
Contributor

@iyastreb iyastreb Sep 18, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Right, I haven't tested with two nodes
Thanks for reducing the change

if (status != UCS_OK) {
return status;
}
Expand Down
Loading