search

NVIDIA / Bobber

Containerized testing of system components that impact AI workload performance
MIT License
14 stars 5 forks source link

Add support for 125KB IO workload #73

Closed joehandzik closed 3 years ago

joehandzik commented 3 years ago

Fixes #71

Signed-Off-By: Joe Handzik jhandzik@nvidia.com

joehandzik commented 3 years ago

If this passes CI, it should be functional - however, I'm realizing I probably need to add to the parser to pull this data out of results. So, we'll see what this looks like and I'll merge myself when it's ready.

roclark commented 3 years ago

Ah, good call, I completely forgot about adding this to the parser. Hopefully it's pretty similar to the output from the other fio tests.

joehandzik commented 3 years ago

@roclark Ok, this is finally ready to go. Package building in GH Actions caught an indenting thing I forgot to check for (boy do I find python indenting rules incredibly arbitrary and subjective). The CI run is indeed shorter now with the tweaks - we can probably tweak further in the future, but it's solidly back down to the previous amount of time (~1.5 hours) but with an additional test in tow. Intentionally not doing baseline comparisons for that particular test yet, but we can add it in as we learn more about what that test is really showing us. Lemme know if you have code issues, and if not, merge away!