512 GB RAM computer for genome assembly

[samanta]

[The question came from a collaborator to me, but I am paraphrasing it to be more readable.]

Would there be enough interest, if we buy a 512G RAM computer and lease it (on a weekly basis) to those requiring high horsepower for sequence assembly and other high performance bioinformatic analysis? The computer will have all necessary programs pre-installed, and we can help pre-load your data, if you mail us DVDs/other storage.

[flxlex]

This could be interesting if you wanted to try out a memory hungry program... How many cpus will there be sharing this memory?

[severin]

Nodes?

When you say 512G RAM computer, does this span multiple nodes?
ie Will it require MPI to run on that much RAM?

[GenoMax]

I assume he means a single server with 512GB RAM (something like R910 from Dell that can have a maximum of 1 TB RAM with 4 x (eight or ten core) Xeon CPU's.

[westerman]

While we (Purdue Genomics) haven't bought into this particular configuration, Purdue's current "community cluster" has the option for 48-core, 512GB memory machines. About $17,000 which includes heat/cool and OS support for 5 years from our central IT. Does not include much disk space -- that is a separate SAN purchase.

So such machines are available at a reasonable price.

[samanta]

Thanks westerman. Is it only for Purdue researchers, or do you allow outsiders to use the machine? If so, at what price?

[westerman]

Our community clusters are only for Purdue researchers (or, of course, collaborators) and are run by the central IT organization. My post was meant to show the cost of a high-memory machine (i.e., not much compared to a sequencer) and not to show off our community clusters. However having been given that opening ...

The 'community cluster' idea is a nice one. When you purchase one or more machines in a cluster then not only do you get access to those machines (runtime maximum of 30 days) but you can also get hold of the other machines in the cluster for 4 hours at a time all without having to worry about heating/cooling and system OS support. The flip-side of that bargain is that when you are not using your machines then other people get to use them for 4 hours at a time.

Another nice thing about our clusters is that we build a new one every year with a 5-year turn-around time. People can trade up and down between the clusters. Thus we are always on the leading edge of hardware while those people at Purdue who do not have the funds for the latest and greatest can get a gently used computer in an older cluster.

[GenoMax]

This is a common model that many of the US universities probably follow. I doubt any allow direct access to anyone not affiliated with the University (unless the access is through a web based open analytical service hosted on the cluster).

Our community clusters are only for Purdue researchers (or, of course, collaborators) and are run by the central IT organization.

Another nice thing about our clusters is that we build a new one every year with a 5-year turn-around time. People can trade up and down between the clusters. Thus we are always on the leading edge of hardware while those people at Purdue who do not have the funds for the latest and greatest can get a gently used computer in an older cluster.