Re: striping?

Dancer wrote:

> No need for raid and no funny business adding cache_dir's.

I assume you are thinking disk-concatenation and not RAID (concatenation
== combining several disks as one large without striping. Usually
possible to add more physical drives when needed). When people say RAID
they commonly think "fault-tolerance", combined with the feature that it
looks like one large disk. (RAID 0 == striping, no fault tolerance)

The cache_dir change in 1.2 is also good news in the fault-tolerance
field. If one cachedrive fails, all you have to do is to remove that
cache_dir from squid.conf and restart Squid. When the disk is replaced
you add the cache_dir line again, run "squid -z" to create the L1/L2
directories and restart Squid. The cache contents on the other disks is
preserved.

Squid currently can't detect malfunctioning drives itself. If a disk
crashes then Squid will perform badly (or even crash) until you remove
the bad cache_dir from squid.conf. If high availablility is important
then this can easily be implemented as a daemon that monitors the
system, rewrites squid.conf if needed and restarts Squid. Another
possibility is to use one of the RAID levels, but it may give poor
performance and requires more disk.

---
Henrik Nordström