SUMMARY Solaris Cachefs

All-

  I got a few replies to my question about using Solaris cachefs on
read-mostly partitions and hit rates. 

  Some folks believed that an 80% hit rate was pretty good, meaning 80%
of the data was coming from local disk and not from the server via the
network. 

  A few responded that cachefs is a wash at best on a LAN because of the
way Solaris cache's NFS data in RAM. For frequently accessed files over
NFS, the files are actually cached into unused RAM. When using cachefs,
the files come from the disk, which is slower than RAM (obviously).
Using cachefs over a WAN (or when you have a slow/overloaded NFS server,
a slow network, ...) would be a win over NFS over a WAN according to
these folks, which makes sense. It also makes sense when the files are
very large (ie they couldn't be cached in available RAM because they are
so big).

  A few others responded claiming that cachefs was the obvious choice
for read-only partitions, and that it is a win all around in all
circumstances. 

  No one knew of a break-even point, however. I'm assuming that the
overhead of checking the consistency and writing the file to cachefs
when it needs update (over and above the time it takes to download the
file via NFS) would make cachefs a loser at some percentage of hits. As
you can see from the summary, though, some folks love it and some folks
think it's loser even at 100% hit rates. 

  Thanks for the input, everyone!

-Eric