performance issues: linux aic7xxx, 29160, Radion IFT-7200

Eric Dantan Rzewnicki rzewnickie at rfa.org
Tue Mar 12 15:07:35 PST 2002


"Justin T. Gibbs" wrote:
> 
> >Hello again,
> 
> I wonder if for some reason we have disconnection disabled?  I get much
> better performance than that talking to a plain old 10K RPM disk.  Can
> you go into aic7xxx.c, search for ahc->user_discenable and print out the
> contents of that variable?

linux-2.4.18/drivers/scsi/aic7xxx/aic7xxx.c
(not-patched, so this is version 6.2.4)
line 4547
ahc->user_discenable = discenable;

same in 2.4.16 ( I guess that's the same version so that would make
sense, of course...)

after patching the 2.4.18 sources for aic7xxx 6.2.5 the file names in
drivers/scsi/aic7xxx/ are different. Which one should I be looking at
there?

> 
> Both drivers negotiate to the same values?

yes, with both 5.2.4 and 6.2.4 the raid and the system disk negotiate to
160.0MB/s

> 
> >Is there something else I can do to try to get better performance with
> >the new driver?
> 
> If I understood why the performance was bad, I'd tell you. 8-)
> Any chance you can rent a SCSI bus analyzer for a day and capture
> some traces?

umm, I'll check with the boss, but I doubt it. Where would a person rent
such a thing?

> 
> >You made a distinction between sequential and non-sequential workloads.
> 
> If you have multple mp3's being written or read from at the same time,
> this will be a non-sequential workload.  If you are only doing one
> at a time, it will be a sequential workload.

Ok, so probably very few people would ever care only about sequential
performance? We will have up to 4 streams of current broadcasts writing
to disk and potentially 'many' users streaming via http or copying via
http or smb from the archive. And for a time, someone will be copying
the past archives off CDs onto the raid via a samba share.

Can hdparm be used to measure non-sequential performance? is there a way
to quantify this?

> 
> --
> Justin

-Eric

To Unsubscribe: send mail to majordomo at FreeBSD.org
with "unsubscribe aic7xxx" in the body of the message




More information about the aic7xxx mailing list