A few things I noticed

[jwm]

Hey Brad, this looks rad. I noticed a couple small things:

• c3.8xl instances have 32 cores, not 64.
• I'm pretty sure regal compute nodes had 32 cores, but I obviously can't check.

Also, re:

During this benchmark, the two limiting factors were CPU usage and IO to the shared filesystem (both disk and network traffic).

I'm not sure network bandwidth was really a factor here. All of these instances have 10ge interfaces, and even though Amazon says the NFS server (a c3.large) has only "medium" network performance, the network graphs are generally well under 1gbit/s. My read of the graphs is that disk performance was a much more significant factor.

[chapmanb]

John; Thanks much for the comments. Apologies, the phrasing is mad confusing there: I'd meant to say 64 total cores and then the memory per machine but agreed that is reads terribly. I'll swap it over to be less confusing.

Good point about network, I'll also ammend that. Thanks again for looking at this and all the helpful comments.