[wplug] Re: SCSI compatibility -- parallel SCSI bus arbitration

[Bryan J. Smith]

On Fri, 2007-06-29 at 15:46 -0400, Michael Skowvron wrote:
> Perhaps you can confirm what I was told about arbitration. I was told by 
> someone that on a parallel SCSI channel, the devices are "serviced" (for 
> lack of a better word right now) in SCSI ID order. In other words, if 
> several disks on the chain are all ready to transfer data to the host, 
> the device with the lowest SCSI ID will allways win or communicate 
> first. On a heavily utilized SCSI bus, this can have the effect of the 
> lower ID devices gaining somewhat of a priority advantage of the higher 
> ID devices.

Good question.

I believe the highest number has priority, with 7 being the host
adapter.  On narrow 8-bit channels, this is the highest.  On wide 16-bit
channels, 7 is still the highest number, and I believe 8-15 map below it
(possibly below 0 as well?).  Can't remember.

In any case, at today's parallel SCSI data transfer rates, you never
want to use more than 3 or, in the worst case, 4 fixed disks per channel
for performance considerations.

If you're deploying new systems, SAS is the way to go.  Parallel SCSI
should only be deployed where and when you have existing capabilities.
I.e., largely replacement or direct augmentation using existing chassis.



-- 
Bryan J. Smith         Professional, Technical Annoyance
mailto:b.j.smith at ieee.org   http://thebs413.blogspot.com
--------------------------------------------------------
        Fission Power:  An Inconvenient Solution