Author: Robert Hyatt
Date: 17:00:45 01/16/03
Go up one level in this thread
On January 16, 2003 at 16:48:18, David Rasmussen wrote: >On January 16, 2003 at 16:36:26, Robert Hyatt wrote: > >> >>Sorry, but maybe you are right, "they are not as bad as I think." >> >>They are _worse_. >> >>:) >> >>Believe me, compared to good SCSI drives, anything IDE simply crawls. I have >>some of the fastest IDE drives around on machines here and they don't even get >>into the same ballpark with U320 15K scsi drives. IE average latency=2ms. 320 >>megabytes/sec burst transfer. Very difficult to even imagine an IDE drive that >>fast. >> > >The real life transfer rate of SCSI disks that I've used is not that much >greater than for good IDE disks. Interface rates are another matter. The >differences can be important in many settings, mostly multi user environments >with many different I/O requests. But on a single user machine running >dedicatedly a chess program, I don't think the difference is huge. Again, I >might be wrong. > >> >>> >>>I know it's slower, but very few people are not using disks comparable to mine, >>>and they don't seem to have a problem. Disks such as yours are not strictly >>>necesary, I would think. Maybe I'm wrong. Also, I still think you are >>>underestimating the performance of current IDE drives. I don't believe that >>>fraction is very small. >> >>We are talking a factor of 2-3-4 at _least_. And when you get an IDE drive >>going >>fast, it saturates the bus and stops everything else. > >What? > >> SCSI doesn't. Try >>working on a >>machine with IDE vs a machine with SCSI and fire up a huge I/O copy. You can't >>notice with SCSI. With IDE you can't use the machine. >> > >You're working with either an old disk or an old controller/motherboard. Modern >IDE incarnations use bus mastering and DMA etc. that makes those statemenst >untrue. And the SCSI harddisks I work with are not unnoticable. I can easily >feel when they work. There is a SunBlade machine I work with, and it has a >10.000RPM SCSI something disk, and the machine is not unaffected when the disk >works. It actually feels slower to work with than a good modern PC with a good >IDE drive. > >>> I have worked on a dell 650 with EIDE drives and a dual xeon 2.2MB. It is a _dog_ when you copy (say) 2 gigabytes from one EIDE drive to another. On the box in my office, as I was setting things up, I did _raw_ disk copies, (in linux dd if=/dev/sdb of=/dev/sdc) and copied 36 gigs of stuff. And I couldn't tell it was running so long as I didn't try to get into the queue to read or write on one of those two drives. There is a _huge_ difference. Particularly with tagged I/O which further offloads stuff from the computer to the controller. Here is a sample of how long it takes to copy the 3-4-5 piece endgame tables to /dev/null from the SCSI drives on this box: crafty% time cat * > /dev/null 0.770u 15.610s 0:037.11 42.1% 0+0k 0+0io 107pf+0w That is 37 seconds elapsed time to read 7.4 gigabytes. Try that on your EIDE machine. BTW I did this while crafty was playing a game on ICC so it might actually run a bit faster by itself... >>>>You might try upping the egtb cache size significantly to avoid some I/O. >>>> >>> >>>To how much, would you say? I tried going from 8MB to 64MB just now, and it >>>didn't help at all. In fact, it was slower. >>> >>>/David >> >>If it is slower you have something else wrong. IE you must be paging. Back off >>the size >>of the normal hash stuff as making the egtb cache bigger should not slow you >>down _ever_. >>unless it begins to cause paging. > >OK, I'll test... > >/David
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.