2D Benchmarking – VM vs Chassis Build

[Scott Miller]

IT wants us to move to VMs. A cost analysis might not pan out, but they want to try optimizing VM builds to suit modeling needs.

First I’ll send a shout out to Chris Goodell for this article: http://www.kleinschmidtgroup.com/ras-post/optimizing-your-computer-for-fast-hec-ras-modeling/

The Corps of Engineers applied benchmarking tests described in this document: https://www.hec.usace.army.mil/software/hec-ras/documentation/RD-51_Benchmarking_2D.pdf

I am not finding where to download the 2D models used for benchmarking. Is anyone familiar with where to get them? It would be great to have standardization, to compare the VMs with physical builds.

It would also be great is you could suggest a 2D stress test, something that could be a standard.

[Luis Partida]

Hi Scott, so the company that I am currently at also has the VM option which most of our numerical modelers use. I would say that the only “problem” with this is that similar to link you posted, is that processor speed is paramount. VM’s and cloud computing are capped at like 2.65 ghz and thats pretty bad even if we can have hundreds of cores available. In my personal experience I had a dell with 32 cores and a processor that turbo’d up to i think 5 ghz and it was SO fast with RAS computations

[Scott Miller]

Nice! My new personal build does 5.5 GHz with the intel 13900k, and it is much faster than my usual workstations.

There are so many factors to account for in comparing where the error comes from in these machines. I’m running a 100-year balanced hydrographs through a low gradient valley with a couple of culverts and foot bridges in the flow. The solution time on Xeon Gold 6242s (2.79 GHz) configured as a 16 core vm are takes 20% to 40% longer than an 18 core Xeon @-2195 (2.3 GHz), varying the number of sockets those cores go through, and has 10x greater volume error. The Core i9 runs in about 60% of the time, but also has about 10x the volume error (1.0% instead of 0.1%). I assume the difference in error on the Core i9 is exactly because it is not a Xeon, but why would the vm have so much more error?

If you know of any guidelines for configuring vms for computation heavy processing, my IT people could benefit.

[Author]

Posts