Here are the official system requirements - 1 GB RAM for CEP2 and 2 GB disk : https://secure.worldcommunitygrid.org/help/viewTopic.do?shortName=minimumreq

Some important background information : The researchers run large molecules that use up to 8 GB RAM on their own linux grid at the lab. They are picking small molecules to send to us that can run on 32-bit compiled code. An improved version of the code is being developed. The current talk is that we will continue to have a 32-bit small version of the improved program but will also have the option of running a 64-bit version for large (8 GB) jobs.

So what does our current code require? I am assuming that molecule picking is not an exact science so some assigned molecules probably exceed the limits. I have no idea how the code enforces limits. If I were doing the coding, I would simply set a 1 GB limit on the largest array and return an error as soon as a molecule tried to exceed it. This would be a loose interpretation of a 1 GB limit since the RAM would include smaller arrays, program code, etc. So a max usage of 1.1 GB would be optimistic but 2 GB seems unreasonably large. If testing showed more than 1.25 GB, I would reduce the max allowed array size.

I do not know how the actual programmers implemented the limit, but if they are like me the limit is probably 1GB plus a bit. After all, virtual memory is supposed to make life easier for programmers by allowing us some slop compared to the really tight coding necessary in the 1960's.

Lawrence

hi Mcgruber,

Here are my 2 cents on the matter: I have also successfully been running CEP2 on a standard (free) RAMDisk partition of just 4GB, with 2 concurrent tasks plus 2 WUs from the other projects. Never ran into trouble. From time to time I would even let S@H and S@H Beta run wild with a WU (a 5th) and still no issues.

I hope this helps.

Cheers!

Are there any other WUs which use more than a few MBs of disk space while crunching? Every time I check the slots folder size of FAAH or FAAHv or SN2S I usually see maybe 10MB or something, and so I usually ignore these as negligible.

Reading your statement, I assume you're tapping 2.2GB or so with the (CEP2)x2 in a worst-case scenario, to which the other WUs add maybe 0.02GB. Given that I don't know exactly the working size max of the CEP2 WUs though, it could also be the case that your CEP2 WUs peg 3.9GB, for a total of maybe 3.95GB conservatively (this seems way too high, though logically still possible).

I suppose this could always change in the future though.

Update:

I configured a script to record my RAMdisk's usage (my /slots folder resides on my RAMdisk) every 15 minutes for 10 hours while one CEP2 (and 3 other different WCG Projects) WU was running.

The highest I got (unfortunately the WU lasted 11:40, while I was only measure for 10:00 of that) was 1.3GB used. Assuming that this usage mostly if not all attributable to the single CEP2 WU, it looks like 1.3GB may be the functional max, so that 1.5GB is a conservative CEP2 WU estimate while crunching.

That said, I'm extrapolating from 40 polling points over a single WU without controlling for the other WCG projects running on my box, so while I'll be running with the 1.5GB number, I caution others against doing so without checking /slot size for yourself.