On 05-Apr-00 Paul Barton-Davis wrote:
> Thanks for the very useful reply. A little minor feedback ...
>>Umm, I don't think so... 80MB/s is an LVD speed. You cannot have both LVD
>>and non-LVD hardware on the same cable in LVD mode. There doesn't appear to
>>be any isolation chip on that MB (and SuperMicro doesn't list one) so it's
>>actually _40_MB/s but that doesn't matter for the purposes here -- the bus
>>is faster than the drive. Besides, you've got too many devices to fit on a
>>proper SE FAST-40 (U2/SE) bus. Again, this is immaterial to speed.
>
> There are 3 SCSI connectors on the motherboard, one U2/LVD-SE, one
> regular old wide SCSI and one Ultra SCSI. The disks are the only
> things connected to the U2/LVD-SE connector. I got the impression from
> the adaptec site that if I set things up this way, I got an isolated
> U2/LVD bus. Is this not true ?
having multiple scsi bus's and segreatting the scsi h/w according to
speed/capabilities, to the different scsi chains, is a very good idea.
That is to say:
all (and only) your LVD devices (regardless of device type) should
go on one of your SE/LVD chains. If there are _any_ SE devices on that
chain, then according to specifications, the adapter must drop down
to SE mode. This will not hurt you so must w/ transfers, as it will
with cable length (40MB SE cable length is _ALOT_ shorter than 80MB
LVD cabling :)
all 40MB SE devices should go on a separate scsi chain (probably
a wide, or ultra adapter, depending upon the devices)
etc, and put all 5mb on their own chain.
the reason is subtle. the scsi bus runs at the speed of teh slowest device.
putting a slow device on a scsi bus will probably affect the performance of
the other scsi devices on that scsi bus.
but none of this is really related to your original question about your
slow scsi performance. your iron's already going as fast as it can. Linux
is basically running it at it's max throughput (as has already been pointed
out). The beauty of LVD, is that you can put alot of LVD devices on a LVD
chain, and they will have little impact on each other (i.e. they can overlap
the IO pretty well).
-Greg
----------------------------------
E-Mail: Gregory Hosler <gregory.hosler@eno.ericsson.se>
Date: 06-Apr-00
Time: 09:47:36
The answer to the meaning of life:
perl -e 'print$i=pack(c5,(41*2),sqrt(7056),(unpack(c,H)-2),oct(115),10); '
----------------------------------
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Fri Apr 07 2000 - 21:00:15 EST