If a site ever did get to the point where one machine
couldn't handle the disk/network I/O, you could
easily split up each domains user dirs across 1(default)
to 63 nodes. 0-9,A-Z,a-z.
Ken Jones
On Thu, 2001-11-08 at 21:11, Doug Clements wrote:
This is offset by the risk of running non-redundant servers.
We have a large NFS store running on FreeBSD (trying to get a Netapp).
vpopmail provides a very effective directory structure which dynamically
accounts for large amounts of domains and users. Also, with a dedicated NFS
server, you can stuff it full of RAM and have it cache most directory
accesses. This takes disk access load off the mail client. It also provides
a central place to back everything up to. Back up one server, and one server
only. If a node dies, you replace it. If the NFS server dies, replace it and
restore a backup of data. We keep 2 IDE drives in the NFS server for
rotating backups, so in a pinch, if the raid fails, we could mount a backup
disk and be back online in minutes.
Of course, with a clustered netapp solution, it makes things so much easier.
They're kinda expensive, though.
From the FreeBSD 4.4 release notes:
A simple hash-based lookup optimization for large directories called dirhash
has been added. Conditional on the UFS_DIRHASH kernel option, it improves
the speed of operations on very large directories at the expense of some
memory.
So if you have tons of memory and still aren't happy with performance, you
can tweak the server to be even faster :)
--Doug
-Original Message-
From: Lu [mailto:[EMAIL PROTECTED]]
Sent: Thursday, November 08, 2001 12:05 PM
To: [EMAIL PROTECTED]
Subject: RE: 5.0 next to 4.10.x
I appreciate everyone's feedback on this.
A couple of last things I need to ask:
Perhaps the downside of using NFS or other shared volume is at a certain
point, the number of directories and files it has to handle will be too
great whereas if I go with separate servers, this is not a problem and
likely an increase in performance.
I guess it all depends on the type of hardware one has but is this
something you take seriously ?
What about scalability ?
Thanks.