[comp.unix.wizards] more on Ultrix 2.0 / Emulex disk throughput

paul@phoenix.Princeton.EDU (Paul Lansky) (09/09/88)

Concerning Emulex QD series Disk controller behaviour under Ultrix 2.0 

> In article <615@hscfvax.harvard.edu>, pavlov@hscfvax.harvard.edu (G.Pavlov) writes:
>    
>     The problem appears to be that DEC decreased the timeout interval on disk
>     requests.  The QD series emulates the MSCP protocol in microcode and does
>     not reply to the cpu quickly enough, thus forcing frequent timeouts.
> 

Can anyone tell me where this timeout occurs? Is it in the uda driver?
its stragegy, start, or interrupt routine?  I have picked through the 
driver a bit and cannot see anything there that would cause this.  The
problem is severe, however.  Under Ultrix 1.1 we were getting a rate of
about 630K bytes per second, and under Ultrix 2.0 this has degraded to
about 250K bytes per second.  I built the kernel from source.
Thank you.  (Both Dec and Emulex seem very tightlipped about this.
The most I could get from DEC was an admission that they shortened the timeout
on HSC to attached disks.  This has nothing to do with our situation, however.)

Paul Lansky
Music Department
Princeton University

paul@phoenix.princeton.edu

chris@mimsy.UUCP (Chris Torek) (09/10/88)

>>In article <615@hscfvax.harvard.edu> pavlov@hscfvax.harvard.edu
>>(G.Pavlov) writes:
>>The problem appears to be that DEC decreased the timeout interval on disk
>>requests. ...

In article <3613@phoenix.Princeton.EDU> paul@phoenix.Princeton.EDU
(Paul Lansky) writes:
>Can anyone tell me where this timeout occurs? Is it in the uda driver?
>its stragegy, start, or interrupt routine?

I rather doubt it.  There is, however, a field in the MSCP `Set Controller
Characteristics' packet called `hosttimo'.  This is the only `timeout' I
know about.  The SCC variant (called, I think, SETCHR in the Ultrix driver)
is structured as:

struct mscpv_sccc {
	u_short sccc_version;		/* MSCP version number */
	u_short sccc_ctlrflags;		/* controller flags */
	u_short sccc_hosttimo;		/* host timeout */
	u_short sccc_usefrac;		/* use fraction */
	long	sccc_time;		/* time and date */
	long	sccc_xxx1;	/* ? */
	long	sccc_errlgfl;	/* ? */
	short	sccc_xxx2;	/* ? */
	short	sccc_copyspd;	/* ? */
};

I would like to know what the `?' fields are for. . . .
-- 
In-Real-Life: Chris Torek, Univ of MD Comp Sci Dept (+1 301 454 7163)
Domain:	chris@mimsy.umd.edu	Path:	uunet!mimsy!chris

pavlov@hscfvax.harvard.edu (G.Pavlov) (09/13/88)

In article <3613@phoenix.Princeton.EDU>, paul@phoenix.Princeton.EDU (Paul Lansky) writes:
> Concerning Emulex QD series Disk controller behaviour under Ultrix 2.0 
> > In article <615@hscfvax.harvard.edu>, pavlov@hscfvax.harvard.edu (G.Pavlov) writes:
> >    
> >   The problem appears to be that DEC decreased the timeout interval on disk
> >   requests......
> 
> Can anyone tell me where this timeout occurs? Is it in the uda driver?
> its stragegy, start, or interrupt routine?  I have picked through the 
> driver a bit and cannot see anything there that would cause this.  The
> problem is severe, however.....
> Thank you.  (Both Dec and Emulex seem very tightlipped about this.

  I was careful to say that this what was relayed to me by Emulex.  I do not
  have source code.  DEC was not helpful.  Emulex was difficult to deal with
  all around.  The entire incident chewed up most of what we saved on purchas-
  ing 3rd party controllers instead of KDA's.  On the other hand, those con-
  trollers let us use Fuji Beagles (M2333K's).

  greg pavlov.