Author: Dann Corbit
Date: 21:40:38 01/22/04
Go up one level in this thread
On January 23, 2004 at 00:12:08, Dann Corbit wrote: >On January 22, 2004 at 23:30:31, Gordon Rattray wrote: > >>On January 22, 2004 at 22:40:25, Robert Hyatt wrote: >> >>>On January 22, 2004 at 22:36:23, margolies,marc wrote: >> >>[snip] >> >>>>And dont put your data to be accessed on a slow hard drive either. A ten >>>>thousand speed rotating SATA (150gb throughput) drive with 78 gigabytes of >>>>storage costs only between 250 and 300 USD. >>> >>>You were doing good until you got to the SATA drive. Throw it away and >>>get a 15K U320 SCSI drive... >> >>I agree that SCSI is fastest. But aren't some of the SATA drives gaining on >>them? A Western Digital Raptor 740 (SATA) can gain an average seek time of 4.5 >>ms. Whereas, a Maxtor Atlas 15k (SCSI) may acheive 3.2 ms. A huge difference?! >> >>And then, the cheaper SATA drives may be put in a RAID config more feasibly in >>terms of cost. So, overall I'm not so sure that SCSI is still so attractive. >>I'm personally thinking of two SATA 10k drives in RAID 1 config. Given that >>I've got an onboard RAID controller, how much would a better SCSI solution cost >>me (2 drives + SCSI controller)? I'm guessing a significant bit more, and not a >>huge performance increase to justify it. > >The best SCSI interface in town: >http://www6.tomshardware.com/storage/20030606/ > >Hotsy-totsy SATA server performance: >http://storagereview.com/articles/200311/20031111WD740GD_4.html > >Hotsy-totsy SCSI ultra320 server performance: >http://storagereview.com/articles/200304/20030429MAS3735_4.html > >Bottom line: >SATA 207 I/Os per second --> highest available performance >SCSI 366 I/Os per second --> highest available performance > >77% improvement. > >SCSI rules in pure performance. SATA rules in price performance. >So if you have to have the ultimate in speed, you need Ultra320 SCSI 15K RPM >drives. >But if you have to have the cheapest I/Os per second, then it is SATA. > >SATA has a problem for database. What happens to a write if you kick the plug >out of the wall in mid-stream? The interface standard does not describe how a >fsync() could be performed reliably. I would be very nervous to store billions >of dollars of data on a large SATA array, unless it had duplicated UPS. Head to head server performance for many disk subsystems: http://storagereview.com/php/benchmark/bench_sort.php
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.