Author: Eugene Nalimov
Date: 11:17:26 01/11/04
Go up one level in this thread
On January 11, 2004 at 14:01:55, Lars Bremer wrote: >On January 10, 2004 at 16:22:25, Robert Hyatt wrote: > >>On January 10, 2004 at 12:13:20, Lars Bremer wrote: >> >>>On January 10, 2004 at 11:51:46, Robert Hyatt wrote: >>> >>>>On January 10, 2004 at 11:30:52, Lars Bremer wrote: >>>> >>>>>Hi, >>>>> >>>>>>I have to buy a new OS :-(( >>>>> >>>>>Win2k does support HT. It handles HT as two different physical processors. WinXP >>>>>can handle these virtual processors. >>>>> >>>>>>The biggest wish is a good Service Pack for Windows NT4 (SP7) with USB, >Directx9 >>>>>>and Hyperthreading support! That's all :-) >>>>> >>>>>It exists and is called Windows XP >>>>> >>>>>>BTW: >>>>>>Bob do you think that SCSI is better as S-ATA (interesting for me if I used on >>>>>>two harddisk 5-pieces for engine-engine matches with ponder = on on dual Xeon. >>>>>>After my first test it seems that S-ATA is just great for tablebases. >>>>> >>>>>Lol, SATA-drives are exactly the same as PATA-drives, there are no differences. >>>>>I count them twice, believe me. >>>>>Normally there is only a special chip at the drive's board to convert parallel >>>>>to serial. >>>>> >>>>>If you want to know which kind of drive is better to store and use tablebases, >>>>>you should read CSS 4/03, where I compared some different hard drives under this >>>>>point of view. >>>>>15k-SCSI was the best, but it was not as fast as one could think, and modern >>>>>ATA-drives are very fast too. >>>> >>>>SCSI drives offer far more than the IDE and IDE-followon SATA drives. >>>> >>>>(1) 320mb/sec burst transfers, double SATA, 2.5X IDE. >>> >>>You talking about the protocol, not about the drives. The fastest SCSI-drives >>>can transfer around 75 MByte/sec, the fastest IDE-drives are close to 60 >>>MByte/sec now. >> >>We are talking two different numbers. I'm talking peak burst transfer from >>on-board cache to main memory. Not sustained data transfer rate for long >>reads. > >Yes, of course. But the peaks have nothing to do with the real life performance. > >>If you think your IDE drive is within 20% of a 15K scsi drive, we can run a >>benchmark to compare them. Every time I have done this, IDE simply gets >>blown out of the water... > >May be you ran the wrong benchmarks. We are talking about tablebases. And here >is the advantage of a SCSI drive not as high as the technical parameters like >seek time and transfer rate would suggest. I tested it with a complete set of >5-men-TBs. I tried following drives: > >drive rpm cache [kB] seek time [ms] transfer rate [Mbyte/s] >U320 Fujitsu MAS3735NP 15000 8192 4,4 70,8 >UDMA6 Samsung SP1604N 7200 2048 10,8 43,5 >UDMA5 Fuj. MPG3204AH-E 7200 2048 11,0 32,7 >UDMA4 Maxtor 33073U4 5400 512 13,4 21,7 > >I let some programs solve some positions and picked the time. A typical result >was: > depth 14 Matt in 13 Matt in 12 >U320 Fujitsu MAS3735NP 00:37 01:24 04:45 >UDMA6 Samsung SP1604N 00:46 01:35 05:08 >UDMA5 Fuj. MPG3204AH-E 00:54 01:45 05:31 >UDMA4 Maxtor 33073U4 01:24 01:45 06:31 > >this was only one program in one position (5R2/5br1/8/4b1B1/8/3k3K/8/8 b - -). >Not too impressive, the SCSI-rocket. > > >>And it doesn't totally hang the system while the >>drives are busy either. >> >> >>> >>>>(2) tagged command queueing >>> >>>tagged command queueing is *not* an SCSI-feature. IBMs IDE hard disk drives can >>>do that since a lot of years. Unfortunatly there is no IDE-driver to handle >>>this. :) >>> >>> >>>>which offloads the "optimizing" stuff from the >>>>I/O request handler and lets the SCSI controller handle multiple requests in >>>>the best possible order, something a request handler can hardly do since disk >>>>drives like to "lie" about their geometry due to various compensation zones. >>>> >>> >>>>(3) run on a SCSI and IDE system side by side. Do something HUGE in terms of >>>>I/O on both. The scsi system will feel perfectly normal. The IDE system >>>>will basically "freeze". >>> >>>It depends. In a server system with a lot of small I/Os, may be. In your >>>computerchess- and desktop-pc, nevermind. >>> >>>>There is little to recommend IDE or SATA except _price_. That is where its >>>>only advantage is seen. >>> >>>So you must be deaf! I never want to have a 15k-SCSI under my desk :) >> >>I have 7 of them and they don't make much noise. The Case fans are far >>noisier. I can barely hear the drives. In fact, I don't notice any >>difference in the 15K drives and the older 10K drives, I have both side-by- >>side. > >May be you are too old to hear the high frequent whistling of an 15k drive ;) >In fact you need a fan for SCSI drives. ATA-drives don't need an additional >cooling. And, your desk must vibrate when your 7 SCSI drives are seeking... > > >> >>> >>>>But I want performance. And for endgame tables, the >>>>faster the better. >>> >>>So you did measure it? What is most important? latency, transfer rate, any >>>other? >> >>Latency _and_ transfer rate. 15K offers the best latency by a factor of 2. >>Transfer rate is next... U320 SCSI wins there as well. > >Well, latency is not the only parameter to describe the seek time. The time to >move the heads between the tracks is a part of seek time too. I do not think >this seek time is the main important factor for TB-access. I tried it with a >test position with the samsung drive, once with acoustic management=fast (seek >time=10,8 ms) and once with acoustic = quiet (seek time = 12,3 ms). There is a >differenc around 10 percent. The time to solve some positions was nearly the >same... > >>>>15K drives are great, U320 15K drives are even better. >>> >>>If you use only one drive there is no difference in speed between U160 and U320. >> >>I disagree, although I have not specifically tested that. But U320 disks >>are out-performing U160 disks in two machines I have sitting side by side. >>IE just doing a disk-to-disk copy of a large file... > >So you use U320 drives with an U160-controller? Come on. If the drives are >different, your "out-performing" says nothing. But this is very off topic. > >greets > >Lars I have one system with SCSI U320 controller and two hard drives plugged in -- U160 and U320. According to specifications those drives have similar seek times. U320 drive consistently outperforms U160 one in disk-heavy activities (TB splitting, compression, copying, verification, etc.) Another good test is program start time in the presence of lot of TBs. I believe Bob reported *very* good start time for his system... Thanks, Eugene
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.