Re: Is noatime supported on Solaris 2.6 and higher?

Suresh Khatry wrote:

> Have 256Mb. The Sparc Server 1000E is also hosting a small web
> site which is not very popular :) Can I have 20GB as cache on this?

Perhaps. Should be ok for lot to medium loads. I would try to add some
memory.

> Planing four disks for cache, 1 for access/cache logs and another
> for swap.state

Fine.

Having a separate disk for swap.state indexes is a good idea from a
stability point of view.

> Do not know of any patches for 2.3S1!!

There are a handful pathes on the "bugs" page
<http://www.squid-cache.org/Versions/v2/2.3/bugs/>. I also have a bit
more than a handful set of patches on my Squid-2.3 patches page
<http://squid.sourceforge.net/hno/patch-2.3.html>

> > e) Use the "fastfs hack" (only if you do not care if the cache
> > filesystem(s) is lost on power-failure..).
>
> How is this done? Please enlighten me.

It is a inofficial hack provided by a sun technican, to enable
non-strict (or actually very loose) filesystem updates. Everything gets
done in memory and then paged out.

Problem is that since there will be many filesystem changes buffered in
memory, chances are close to 100% that fsck cannot recover the
filesystem after a powerloss.

The DiskSuite fs-logging capability is very similar to this but without
compromising the integrety of the filesystem in case of powerloss, if
you can put the fs-transaction-log on a drive which is otherwise mostly
idle.

--
Henrik Nordstrom
Squid hacker