Back to the RAID question

More research on the RAID issue has me more discombobulated.  For
example, while theory says read speed on a RAID-5 setup can beat that of
a RAID-1, simply because of more spindles, any real benchmarks I can
find indicate that this isn't the case.  The only hard numbers I could
find on the 'Net (done on several models of Solaris machines) indicate
that the two are similar in read speed, and that RAID-5 falls off on
WRITE.

Now, one difference is that the configuration in the tests was RAID-1 at
3-3 vs. RAID-5 at 1-3, meaning the RAID-1 machine had six drives, so
that may be a contributing factor.

IBM's sales line, on the other hand, contends that RAID-1 is faster on
reads, and RAID-5 is faster on writes.  When I asked for any sort of
corroboration, the guy on the phone line said he didn't have any, but
that's what the "expert" told him.  When I asked how the expert knew, it
was clear that this guy thought I was wasting his time.

Anyway, the RAID-5 solution is significantly cheaper as the number of
drives goes up (5 36GB drives on a RAID-5 is $500 less than the
equivalent 2 146GB or 4 73GB drives on RAID-1), and is also cheaper for
incremental disk additions.  I think that unless someone has some hard
evidence that RAID-5 will significantly under-perform RAID-1, then I
think that's the direction to go.

Joe