On 4/12/01 at 02:02, ie9t-sbgk@asahi-net.or.jp (Takeshi Shibagaki) wrote:

> Hi,
> 
> From: "Michael G. Schabert" <mikeride@mac.com>
> Subject: Re: installing NetBSD on an IDE HD
> Date: Mon, 3 Dec 2001 11:17:54 -0500
> Message-ID: <p05100301b8315470a5b0@[192.168.1.4]>
> 
> mikeride> That benchmark is doing one thing at a time. IDE is perfectly fine 
> mikeride> for that. The problem with IDE is that it does not support
concurrent 
> mikeride> I/O. This is quite important for many un*x tasks, particularly if 
> mikeride> you're actually going to be touching swap space during use. As Un*x 
> mikeride> is designed as a multi-user, multi-tasking OS, concurrent I/O can 
> mikeride> play a crucial role in performance. This is what the original poster 
> mikeride> was getting at.
> 
> What do you call for "concurrent I/O".
> 
> mikeride> It *used* to be that IDE was slower primarily because it was so CPU 
> mikeride> intensive. In today's computers, however, the CPU's just sitting 
> mikeride> there doing nothing anyway, since they're so darn fast. Today, there 
> mikeride> are 2 main diifferences between SCSI and IDE.
> mikeride> 
> mikeride> 1) The aforementioned concurrent I/O
> mikeride> 
> mikeride> 2) SCSI drives are better quality than IDE. They are designed for 
> mikeride> server applications and are built to much higher specs. As my word
is 
> mikeride> irrelevant ;-), a quick test would be to go to the website of any 
> mikeride> manufacturer that makes both (IBM for example) and look at the specs 
> mikeride> such as MTBF, duty cycle, etc. There is generally a fairly high 
> mikeride> discrepancy between the IDE offerings and the SCSI offerings.
> 
> It is a general theory, but in this case, I don't think so.
> In port-mac68k, SCSI is pseudo DMA after all, so CPU time in SCSI
> is same to IDE almostly. port-mac68k isn't port-i368(for example).
> 
> And 2) is also a general theory. Today, IDE driver is more
> popular than SCSI driver. So I think both are equal quality.
> 
> # I allow SCSI hardware is better than IDE one, but case by case.
> 

I have made some testing with my ide and scsi drives:

Machine: performa 630 48 MB RAM 1.5.2-GENERIC

Bonnie results
______________________________________________________________________________

IDE Samsung WNR-31601A 1.6 GB 
       
              -------Sequential Output-------- ---Sequential Input-- --Random--
              -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
Machine    MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU  /sec %CPU
          100   466 93.7   839 49.0   504 54.2   421 94.9  1060 87.7  22.5 23.4
            

SCSI Fireball 540 MB            
            
              -------Sequential Output-------- ---Sequential Input-- --Random--
              -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
Machine    MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU  /sec %CPU
          100   515 98.4  1162 80.5   481 53.7   475 99.4  1354 83.8  26.9 20.1


ByteBenchmark results
______________________________________________________________________________

IDE Samsung WNR-31601A 1.6 GB

File Read  (10 seconds)                    6280.0 KBps  (10 secs, 6 samples)
File Write (10 seconds)                     693.0 KBps  (10 secs, 6 samples)
File Copy  (10 seconds)                     454.0 KBps  (10 secs, 6 samples)
File Read  (30 seconds)                    6526.0 KBps  (30 secs, 6 samples)
File Write (30 seconds)                     666.0 KBps  (30 secs, 6 samples)
File Copy  (30 seconds)                     477.0 KBps  (30 secs, 6 samples)

SCSI Fireball 540 MB

File Read  (10 seconds)                   11666.0 KBps  (10 secs, 6 samples)
File Write (10 seconds)                    1000.0 KBps  (10 secs, 6 samples)
File Copy  (10 seconds)                     626.0 KBps  (10 secs, 6 samples)
File Read  (30 seconds)                   12115.0 KBps  (30 secs, 6 samples)
File Write (30 seconds)                    1000.0 KBps  (30 secs, 6 samples)
File Copy  (30 seconds)                     713.0 KBps  (30 secs, 6 samples)


The results should be normal, but I think my IDE drive is better than the scsi,
in MacOS side the IDE drive has a better performance than the scsi, at least
norton utilities shows that.