mirrored file systems

[Larry McVoy]

In article <47 at xyzzy.UUCP> goudreau at larrybud.dg.com (Bob Goudreau) writes:
>In article <4882 at amelia.nas.nasa.gov> truesdel at sun217..nas.nasa.gov (David A. Truesdell) writes:
>>In addition, a mirrored filesystem won't help your I/O throughput.
>
>Actually, that depends on the system's read/write ratio.  Writes to
>a mirrored file system obviously must result in I/O to each (valid)
>side of the mirror.  But reads, on the other hand, only need to
>find *one* good copy.  An intelligently implemented mirrored file
>system can increase I/O throughput by distributing read requests
>evenly among the copies, thus leaving the unused sides of the
>mirror free to service other incoming read requests.

Good point.  Here's something else to consider though, when thinking about
this sort of thing.  When you speak of the  read/write ratio, be careful
to tell me which side of the buffer cache you mean.  The read/write
ratio on the system call side is about 4:1, but on the other side I've
found it's much more like 1:4. In other words, the cache caches lots of
reads, but the writes end up going through (most of the write traffic
I've seen are directory updates since they are synchronous).  Just a 
thought...
---
What I say is my opinion.  I am not paid to speak for Sun, I'm paid to hack.
    Besides, I frequently read news when I'm drjhgunghc, err, um, drunk.
Larry McVoy, Sun Microsystems     (415) 336-7627       ...!sun!lm or lm at sun.com