Author: James R Grinter Date: To: Chris Thompson, John Henders CC: exim-users Old-Topics: Re: Local delivery problems on FreeBSD Subject: Large directories (was Re: Local delivery problems on FreeBSD)
On Thu 28 Nov, 1996, Chris Thompson <cet1@???> wrote: >The efficiency of access to single files in very large directories tends
>to be getting better in modern systems. Both Solaris 2 and the Network
>Appliance fileserver seem to have coped OK with our 6000+ entry /var/mail
>on cus.cam.ac.uk.
NetApp file systems have a trade off: when you set up the filesystem
you'd have configured the maximum number of files that can be in
one directory, and it won't let you have any more (I can't remember
what the default number is).
For Solaris 2, it still has the UFS/FFS characteristics, though I
get the impression that it's better at sizing inode caches to cope.
(The only filesystem that genuinely copes with millions of directory
entries, to my knowledge, is SGI's XFS).
But, this 'large directory problem' reminds me that another feature
I'd like to see incorporated is splitting out /var/spool/exim/input/
(or similarly named, ie the queue directory) into subdirectories,
Either using some hashing scheme, or making use of the most frequently
changing part of the Message-Id.
