Re: FastSCSI on WarpEngine
On Tue, 2003-03-04 at 16:55, Ingo Juergensmann wrote:
> On Tue, Mar 04, 2003 at 02:23:07PM +0100, Richard Zidlicky wrote:
>
> > > I would like to see around 4-6 MB/s coming out of the disks on that machine,
> > > so I wonder how to achieve this. Maybe the mentioned change in the source
> > > code from 50 to 25 would be a way. Instead of facing the frontier and risk
> > > data corruption with 10 MHz transfer a value of 37 could mean 7.5 MHz
> > > transfers... don't know whether this might work or not...
> > Hm.. I get only 3.3 MB/s peak performance and perhaps 2 MB/s realistic
> > performance on my 68060 with an IDE disk, would be interesting to see
> > a few more datapoints. Iirc my IDE is on something like a 10 MHz bus.
>
> Yeah, but it's IDE and when I saw IDE performance in crest (when I set it
> up), it was awful... IDE speed was around 1-1.5 MB/s on the internal IDE bus
> whereas SCSI was twice as fast with a really old Quantum LPS105...
>
> > I remember a 68040 40 MHz had so little reserves im memory bandwidth
> > and CPU power that even io buffered in RAM (not in disk buffer) had
> > a maximum throughput of about 5.6 MB/s.
>
> Well, when I would be able to reboot arrakis now, I could have a look at
> that, but AFAIR memory bandwidth was much higher, something around 40
> MB/s... (I usually tested this with bustest program on AmigaOS, see Aminet)
>
> What could be is that memory bandwidth is limited under Linux due to some
> restrains of memory protection and context switches or something like this,
> which doesn't occur under AmigaOS...
I have a Blizzard 1260 with a 68060 @ 50 MHz here, and doing a quick
'hdparm -T' (to test buffer-cache reads) gives me the following results:
/dev/hda:
Timing buffer-cache reads: 128 MB in 9.68 seconds = 13.22 MB/sec
/dev/sda:
Timing buffer-cache reads: 128 MB in 9.67 seconds = 13.24 MB/sec
/dev/sdb:
Timing buffer-cache reads: 128 MB in 9.69 seconds = 13.21 MB/sec
Using 'hdparm -t' (to test device read timings) yields:
/dev/hda:
Timing buffered disk reads: 64 MB in 50.48 seconds = 1.27 MB/sec
This is on a 3.2 GB 2.5 inch IBM DTCA-23240 IDE disk using the internal
A1200 IDE interface. Which I don't even find _that_ bad for a PIO mode 0
controller ;-)
/dev/sda:
Timing buffered disk reads: 64 MB in 23.33 seconds = 2.74 MB/sec
This is an old 2.1 GB Quantum Fireball TM2110S disk on my Blizzard IV
SCSI kit (which is a Symbios Logic 53C9x-2 chip @ 40 MHz), period 100ns,
10 MHz FAST synchronous)
/dev/sdb:
Timing buffered disk reads: 64 MB in 16.41 seconds = 3.90 MB/sec
This is a 7200 RPM Seagate Baracuda ST32550W disk on the same controller
with the same settings.
Summarising: I don't think you'll ever get 4-6 MB/s with a single disk.
You don't even fill up the 5 MHz bandwidth currently it seems, so I
don't think it will get much better.
Of course, I don't know how good or bad the Blizzard 1260 design is, I
heard that its memory bus is clocked really slow (25 MHz?) (that might
include the SCSI controller).
Kind regards,
Kars.
Reply to: