From: jreed@appliedtheory.com
Date: Thu Aug 01 2002 - 10:14:38 EDT
I'm trying to set up a striped mirrorset (stripe across 5 mirrors)
on an esa10000 with hsz70 controllers. The array configuration
manual says striped mirrorsets provide very high performance. However,
when we actually create the stripe, it says:
"This storageset is configured with more than one disk per port. This
causes a degradation in performance, and a higher risk of data
inaccessiblity."
The physical layout looks like this:
m1 m2 m3 m4 m5
m1 m2 m3 m4 m5
with stripe0 = m1 m2 m3 m4 m5.
When the manual shows mirrorsets in its explanation of a striped mirror,
it shows them as consisting of two disks on the same bus, ie:
m1 m2 m3 m4 m5
m1 m2 m3 m4 m5
even though at other places in the manual it shows creating mirrorsets
across 2 buses for reliability. It doesn't make sense to me to make
mirrors of 2 disks on the same bus, as that means if the bus goes the
mirror is gone, and if that mirror is part of a 1+0 stripe, the stripe
is history. However, if the first config I describe, with mirrors across
two buses, truly degrades performance in the striped mirror then it
doesn't make sense to build a striped mirror at all.
Does anyone know anything authoritative about this, or have any real
date on performace vs. degradation? This relates to a question I asked
awhile ago about how oracle wants the database on one huge striped mirror,
which really made no sense on this kind of array. Replies were very mixed
about that config anyway, I'll summarize soon. I'm starting to think,
however, that striped mirrors larger than 3 mirrorsets are not adviseable
on these arrays - I think that:
m1 m2 m3
m1 m2 m3
would work across the six busses without the degradation they are concerned
with, but nothing larger will work. ???
TIA,
-- Judith Reed jreed@appliedtheory.com (315) 453-2912 x5835
This archive was generated by hypermail 2.1.7 : Sat Apr 12 2008 - 10:48:48 EDT