ucx icon indicating copy to clipboard operation
ucx copied to clipboard

UCS/SYS/TOPO: Added options to set distance between devices.

Open rakhmets opened this issue 1 year ago • 3 comments

What

Added options to set estimated latency and bandwidth values according to the distance within the sysfs device tree. Moved platform specific values to ucx.conf file.

rakhmets avatar Jun 10 '24 15:06 rakhmets

/azp run

brminich avatar Jun 11 '24 11:06 brminich

Azure Pipelines successfully started running 4 pipeline(s).

azure-pipelines[bot] avatar Jun 11 '24 11:06 azure-pipelines[bot]

Another idea that came to my mind is that maybe we don't need so many environment variables that do the same thing but for different cases. Maybe better to define something like DISTANCE_BW=numa:17000MBs and DISTANCE_LAT=sys:500ns using key-value config type as it is done for RNDV inter/intra thresh?

I understand that there can be complications with PCI case since it contains also BW coefficient, but I think it can be a separate config value.

ivankochin avatar Jun 11 '24 13:06 ivankochin

/azp run

rakhmets avatar Aug 20 '24 15:08 rakhmets

Azure Pipelines successfully started running 5 pipeline(s).

azure-pipelines[bot] avatar Aug 20 '24 15:08 azure-pipelines[bot]

@rakhmets pls squash

yosefe avatar Aug 26 '24 13:08 yosefe

/azp run UCX PR

rakhmets avatar Aug 28 '24 07:08 rakhmets

Azure Pipelines successfully started running 1 pipeline(s).

azure-pipelines[bot] avatar Aug 28 '24 07:08 azure-pipelines[bot]