Re: surprising results (to me.)
Ideally, test the actual code you'll be running.
That's usually a lot of work, and can be expensive.
One of the reasons I sometimes go ballistic on these boards, is that on the handful of test we've had the chance to run we've found that the performance of the P4 is being grossly overrated, just about everywhere, by just about everyone.
And this misrepresentation of the P4's performance has led to a millions of very bad purchase decisions. It's a big swindle, on the level of Enron in total dollars, but since each case (each CPU sale) amounts to only a few hundred dollars, it gets ignored.
I buy, or have built, workstations and servers. The most recent servers I've speced have had large storage capacity and computation requirements, but low transaction requirements (a handful of researchers hammering on very large tabular and spatial data sets). Most recently, I've been specing systems with Antec's 1480 case (with redundant PS), dual Athlon MP 2000+ chips, Asus's A7M266-D motherboard, 2-gig or 3-gig of RAM, 3Ware's Escalade RAID controller, and 9 WD 80gig 7200 RPM drives with 8meg cache. That leaves me with a cold spare, a hot spare, and 560 gig of available RAID protected storage. These systems go for $4,500 to $5,000, depending on memory. They're cheap enough to just buy a spare system, test it, and leave it in the box. A slower (in CPU) Dell, with half the storage, is $15,000. I have these systems built by a local VAR.
For a transaction oriented system (which haven't been the most recent procurements), I've been specing similar boxes, but with the Adaptec 3200S SCSI RAID controller and Seagate 10K or 15K drives, which triples the cost of the disk, and doubles the cost of the server. |
