Volatile performance with c220g5 machines

[Florian Suri-Payer]

I was running some experiments last Friday and today on the Wisconsin cluster (c220g5) machines. In both cases I used identical code and experiment setups but I (consistently) measured significant performance differences (2x). 

Somehow there seems to be a large performance volatility between c220g5 machines, and I must've ended up with an "unlucky" cluster today somehow. I am quite certain it was not a config issue on my end -- I just switched to Utah (m510) and everything runs the way I'd expect again.
I unfortunately can't provide much more insight, but I thought this was odd and you may want to look into the machines/network.

Best,

Florian