Non-intuitive gains in performance

[bill davidsen]

In article <yl66ijw6y7.fsf at windlord.stanford.edu>,
Russ Allbery  <rra at stanford.edu> wrote:
| 
| Bill Davidsen <davidsen at tmr.com> writes:

| > So I set the metacycbuff SEQUENTIAL as advertised, and the artwrite time
| > didn't go down a bit. However, the hiswrite time dropped by some 40%! 
| > Does anyone have a thought on this? It's nice that I really helped the
| > performance, but I lack any idea why it works this way.
| 
| Is history on the same drives as the storage?  What's the disk
| configuration?

  History and active are on a RAID-1 array of 9GB 10k rpm drives, for
better read performance. Storage is on RAID-5 36GB 7200 rpm drives.
Looking with vmstat2 I see 30-40 ops/sec, which shouldn't be much of a
load. Expire peaks ~170 ops/sec on the history array.

  I don't see why a change to the CNFS write order would hit the
history, but I'll take any gains I can get.
-- 
bill davidsen <davidsen at tmr.com>
  CTO, TMR Associates, Inc
Doing interesting things with little computers since 1979.