[SunRescue] Question about SSAs

[Patrick Giagnocavo]

On Sat, May 05, 2001 at 12:38:23PM -0700, David Passmore wrote:
> On Fri, May 04, 2001 at 09:01:39PM -0400, Dave McGuire wrote:
> 
> >   On the subject of FCAL...I'm about to light up an A5100 on an Ultra30
> > with a QLA2100.  I've never managed an FCAL-based system before, and
> > the I/O performance of this system is going to be rather
> > critical...can anyone offer any tips for performance tuning and/or
> > monitoring?  The system is running Solaris7, with which I'm also
> > somewhat unfamiliar...I'm a BSD guy.
> 
> I did an evaluation of A5200s vs. EMC Symmetrix last year; 10 fully
> populated A5200s, 220 18GB spindles, vs 96 36GB spindles in the Symm with
> 8GB of cache.

Then you sound like the guy to ask:

Have you tested performance when the disk fills up?

Physical media write speeds on a disk vary depending on which part of the
disk is written to.  On SCSI and EIDE/ATA drives, you have a logical block
address (dunno if terminology is same on both interfaces) scheme where 0
denotes the first byte on the outer diameter (OD) of the disk, and the last
byte in the address scheme is on the inner diameter of the disk (ID).

Since the disk is spinning at a fixed RPM, read/write speeds are higher on
the OD than the ID.

According to one guy I know, large arrays like EMC etc. NEVER use more than
half the disk, since performance begins to degrade quite a bit - not only do
you use the slower ID part of the disk, but the head has to travel more,
resulting in slower seeks.  Some makers use the "middle" part of the disk in
an effort to further reduce seeks.

It would be interesting to see what happens between two competing arrays
when disks are nearly full - would reveal flaws in the array design rather
than differences in disk performance.

Cordially

Patrick Giagnocavo
patrick at zill.net