From: leop@percy.rain.com (Leo) Subject: Re: SCSI buyers information Date: 8 Aug 1993 20:46:29 GMT
davidsen@sixhub.UUCP (Wm E. Davidsen Jr) writes:
>In article <PCG.93Aug1215110@decb.aber.ac.uk> pcg@aber.ac.uk (Piercarlo Grandi) writes:
>| Only if miraculously enough you manage to run *both* disks at their
>| rated maximum speed *at exactly the same time* for any length of time.
>| This is not very realistic. Actually, it is incredibly optimistic. The
>| idea that a disk that can deliver 2-4MB from the platter is going to
>| deliver it over the bus sustained is peculiar.
> my experience is that you are correct in practice as well as
>theory. Using a 1542B vs. 1742, on the same machine (EISA/VESA
>486-66) showed almost no difference in performance, even when
>doing things like memory buffered copy from one disk to another.
>When you get more than two disks, assuming heavy load, then you
>can get some performance problems, but that's a really heavy
>load, and I'm not sure if Linux will generate i/o on more than
>two disks (since I don't have source handy).
There is a similar discussion taking place in
comp.sys.ibm.pc.hardware. The benchmarks in the message below seem to
indicate quite a difference. Is it due to the difference between
drivers on Linux and 386BSD?
From: andrew@werple.apana.org.au (Andrew Herbert)
Date: 8 Aug 1993 19:25:45 +1000
Newsgroups: comp.sys.ibm.pc.hardware,comp.unix.pc-clone.32bit,comp.periphs.scsi
Subject: adaptec 1542B vs 1742A (was Re: Adaptec vs Ultrastor)
bob@obiwan.uucp (Bob Willcox) writes:
>(this system) running 386bsd. Reading from the raw disk device
>does seem to be faster with this system vs. one of my others that
>has a 1542B. This is, of course, comparing apples to oranges but
...
True, but I found the same thing when I moved from a 1542B to a 1742A,
keeping everything else constant. FYI, I've also included some stats from a
barracuda. The following stats are all for a i486/33 EISA system. The
results were obtained using iozone with a 16 MB test file and 1024 byte
transfers, under netbsd 0.8a, and represent bytes/sec throughput (through a
FFS filesystem with 8 KB blocks & 1 KB frags).
controller/drive combo write read
aha1542b + CDC wren 6 (scsi 1) 210241 519955
aha1742a + CDC wren 6 (scsi 1) 435019 1331525
aha1742a + Seagate Barracuda 1 (scsi 2) 1878047 2088450
Note the barra figures are on a lightly loaded system - in single user mode
one gets about 1.9 MB/s and 2.6 MB/s write & read performance respectively. :-)
Andrew