Author: Robert Hyatt
Date: 17:41:06 02/21/03
Go up one level in this thread
On February 21, 2003 at 13:48:13, Matt Taylor wrote: >On February 21, 2003 at 09:51:34, Robert Hyatt wrote: > >>On February 21, 2003 at 05:31:35, Matt Taylor wrote: >> >>>On February 20, 2003 at 18:23:22, Robert Hyatt wrote: >>> >>><snip> >>>>My cpus don't run blazing hot, although there is another significant performance >>>>difference >>>>in our boxes. I can copy 8 gigs of data in 30 seconds. That is important for >>>>chess endgames >>>>with tables. I don't think you can get 1/10th of that speed with IDE drives of >>>>any kind. >>>>But that is yet a different subject. >>><snip> >>> >>>(2) Western Digital 80GB 8MB cache 7200 rpm >>>3ware 7000-2 RAID controller >>> >>>Buffered Read: 94 MB/sec >>>Sequential Read: 49 MB/sec >>>Random Read: 6 MB/sec >>> >>>Buffered Write: 87 MB/sec >>>Sequential Write: 58 MB/sec >>>Random Write: 12 MB/sec >>> >>>This would be 1.4 GB in 30 seconds. Nowhere near 8 GB, but it's more than 10%. >>>Also, this was done on an NT box using NTFS, and NTFS is notoriously slow. >>>Seeking all over a 7200 rpm drive really cripples the "sequential" speed. >>> >>>That is the low-end model RAID controller in 2-disk RAID-0. 8-disk RAID-0 is >>>pretty fast. They claim speeds up near 600 or 700 MB/sec in which case it would >>>be copying 18 GB in 30 seconds. The 7500-8 costs something like $350 or $250, I >>>believe. That array would be about $1,000. The 8500-8 is what I would use if >>>money were no object. >>> >>>-Matt >> >>I've got an 8-drive IDE raid array here we are playing with for backups, and >>it is _not_ fast. This is an 8 x 250gig device and no, I don't remember who >>made it as it isn't ours (ours = CIS) it was an evaluation device from our >>IT folks up the street. However, the random speed is more important than the >>sequential read for tablebases, and the 15K SCSI drives have 1/2 the latency >>of the best IDE drives so far that I have seen (7200 RPM). I'll run the 8 gig >>copy on this array monday to get a number, but when I tried it last time it >>took over 5 minutes total. Which is where I got my 10x number. >> >>Beware of claimed speeds also, as that is typically drive buffer to memory, >>which is not the same as copying large files where buffer means zero. > >Throughput listed for Barracuda V drives is listed as 72 MB/sec. > >Here is benchmark data for the Escalade in particular: >http://www.3ware.com/products/benchmarks.asp > >For some reason, I was remembering it as 600-700 MB/sec. It is much lower, but >it still comes to around 6 GB over 30 sec sequential read speed, and that's on a >5400 rpm drive. The Barracuda V drives would abosolutely fly. > >10K rpm ATA drives are coming. Western Digital will soon release a line of 10K >rpm Serial ATA drives, basically a SCSI drive on a Serial ATA interface. I think >rpms is really the -major- seperating factor between SCSI and ATA right now. A >good ATA RAID controller will do everything the SCSI disk already does. > >-Matt Note that 10K won't affect bandwidth whatsoever, it just changes two things. Cuts latency to 2ms, but reduces the drive data density by 2x so that you can't get the ultra-fat drives like the 146gig drives I bought for the new ftp platform. The _main_ thing that separates SCSI and IDE, besides bus efficiency, is "tagged command queueing". So that you can fire off dozens of reads to every drive on a controller and overlap the reads and seeks like crazy. IDE won't do that. The best way to see the difference is to take a machine with two scsi drives (one channel) and put it side by side with a machine with two IDE drives on one IDE channel. Do some sort of huge disk copy from the second disk to /dev/null (unix) on both machines. Now try to do something on the _other_ drive. Or even try to do something on the IDE machine at all... The IDE machine will feel sluggish. The SCSI machine will feel just like it always does. That highlights the difference in a startling way...
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.