Re: Sun SS10 model 30 Performance

From: John Grana <jjg_at_pt.com>
Date: Wed, 21 Oct 1992 23:29:03 GMT
Message-ID: <1992Oct21.232903.28102_at_pt.com>


In article <bartho.36.719499953_at_cc.und.ac.za> bartho_at_cc.und.ac.za (Ray Bartho) writes:
>We are currently benchmarking an Oracle database application on a Sun SS10
>Model 30 with SunOs 4.1.3 (v2), having performed the same tests on an HP
>9000/817 and an IBM RS 6000 Model 350.
>
>The impression was not favourable since the database import was two and a
>half times slower on the Sun (5hrs vs 2hrs on the HP and IBM). Other
>database jobs doing only reading of the database perform roughly the same or
>better, but it appears that write access to disk is slow.
>
>The IO Benchmark program IOZONE by Bill Norcott confirms this slowness with
>an average of 1.7 Megabytes per second no matter what file size or block
>size is used. It is almost as if there is no (or very little) cacheing
>taking place. We have looked through the SunOS kernel configuration, but
>cannot see anything that relates to the problem. Increased USERS to 32, but
>this also had no effect.

I think that the 1.7MB/Sec is pretty good (on writes). The disk drive can probably only write the actual "spinning media" at around 2.1 or so MB/sec. Also, notice how even small I/O requests (512 to 2K) go pretty fast. I have been doing lots of SCSI benchmarking the past few days on an SS10 and have been pretty happy ('course I would since our SBus H.A. (SBS430) is getting around 2 MB/Sec!) For a reference, run the IOZONE benchmark on a SS2. You might be suprised, esp. at small requests (300-500KB/sec). It seems the old sweet spot at 8K is now spread over most all I/O sizes (at least on the SS10 running 4.1.3).

-- 
___________________________________________________________________________
|John Grana, Performance Technologies Incorporated		jjg_at_pt.com|
|315 Science Parkway, Rochester, New York 14620           uupsi!ptsys1!jjg|
|Phone: (716) 256-0200   Fax: (716) 256-0791                              |
Received on Thu Oct 22 1992 - 00:29:03 CET

Original text of this message