[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Condor-users] Best practices for CPU architecture optimizations?



We're discussing how best to set up for steering jobs optimized for and/or requiring AVX extensions[1]. I'm hoping someone else has a best practices guide on expressing and then ranking or requiring CPU flags in classad's.

The approach I'm thinking of is setting a CpuFlags ClassAd to:

grep flags /proc/cpuinfo | head -1 | cut -d: -f2

...so one could use a Rank or Requirements expression of:

stringListMember("sse2",CpuFlags," ")

...for SSE2 jobs, for instance. Is that what others are doing already, or is there a better practice out there from the folks that have been doing this longer?

[1] http://en.wikipedia.org/wiki/Advanced_Vector_Extensions