Large dirs [Was: Re: Clustering and replication]
janne.peltonen at helsinki.fi
Wed Jan 31 01:32:38 EST 2007
On Tue, Jan 30, 2007 at 02:27:24PM -0800, Tom Samplonius wrote:
> It could be filesystem and kernel related. ext2/3 has historically
> not been good at this, but is better now. ufs is actually pretty
> good, but mainly because BSD kernels cache so much vnode data, it
> does not matter if the directory is hashed on disk or not.
> And you always want "noatime" enabled on your Cyrus spool, no matter
> the filesystem or OS, or version of Cyrus.
I've got a Solaris 8 ufs with noatime. And this was a one-time one-user
problem during extremely high server load. So probably not worth
investigating much more closely.
More information about the Info-cyrus