RCF Linux Procurement Plan


We have now received input from all of the experiments on their preferred purchase configuration for the FY ’03 Linux farm procurement within (and slightly beyond) the constraints that exist. I will describe here what RCF proposes to do and why. I will be away until Dec 30 so you have until then to think about whether there are problems with this plan that we have not considered. We will have a meeting that week to discuss such thoughts before we actually issue a bid package. The plan is as follows:

We will purchase only 1 type of node: 1-u, 2.4 GHz dual P4, 1 GB mem, 2 x 120 GB IDE disk

Since all 4 qualified vendors offer it, and the estimate of their prices are very similar, we are highly confidence of getting a very good price on such a large purchase (>240). Also for the longer term it is simpler to install, configure, manage and maintain a single type of machine.
For those experiments for which this is not seen as a disk heavy enough solution, we will support them in using some of their funds to add disk to existing 2-u systems in the facility; for PHOBOS this means their option (3) which very minimally meets the 50 TB criterion. This also applies to PHENIX, which asked for ten 2-u 4 SCSI disk nodes.

Also for PHENIX, which asked for SCSI based 1-u’s we refer to Jerome’s measurement of SCSI versus IDE disk performance* in which he saw significant differences in write performance but very little difference in read performance, which is the way in which we expect RCF users to stress their I/O. Any performance issue should be, at least partially, off set by the fact that the cost saving from going to IDE is likely to enable the purchase of ~10 additional machines.

This solution physically fits, with ~1.5 racks to spare, in the available Linux area, allowing for the possibility of increasing the scale of the purchase, as we have done in the past, when we get a good price. Note that if a 2-u IDE solution did exist for PHOBOS it would totally saturate the available space, exceeding their 4,4,2,1 share of that space and interfering with the expansion by other experiments if the pricing turns out to be favorable. The moving of ATLAS, while discussed, would be premature since additional space for RCF/ACF will not become available until next summer and so a double move would almost certainly be necessary. This would more than double the cost and effort and also double the disruption and risk to ATLAS with no benefit for them.










* http://www.star.bnl.gov/STAR/comp/sofi/iotest/SCSI_IDE/info.html

Bruce G. Gibbard 12/20/02