From owner-freebsd-hackers Sun Feb 18 20:13:14 1996 Return-Path: owner-hackers Received: (from root@localhost) by freefall.freebsd.org (8.7.3/8.7.3) id UAA25769 for hackers-outgoing; Sun, 18 Feb 1996 20:13:14 -0800 (PST) Received: from brasil.moneng.mei.com (brasil.moneng.mei.com [151.186.109.160]) by freefall.freebsd.org (8.7.3/8.7.3) with ESMTP id UAA25764 for ; Sun, 18 Feb 1996 20:13:07 -0800 (PST) Received: (from jgreco@localhost) by brasil.moneng.mei.com (8.7.Beta.1/8.7.Beta.1) id WAA08332; Sun, 18 Feb 1996 22:04:22 -0600 From: Joe Greco Message-Id: <199602190404.WAA08332@brasil.moneng.mei.com> Subject: Re: Some SERIOUS NFS usage - advice needed To: msmith@atrad.adelaide.edu.au (Michael Smith) Date: Sun, 18 Feb 1996 22:04:22 -0600 (CST) Cc: rashid@rk.ios.com, hackers@FreeBSD.ORG In-Reply-To: <199602190213.MAA07762@genesis.atrad.adelaide.edu.au> from "Michael Smith" at Feb 19, 96 12:43:36 pm X-Mailer: ELM [version 2.4 PL24] Content-Type: text Sender: owner-hackers@FreeBSD.ORG Precedence: bulk > Rashid Karimov stands accused of saying: > > > > We have this idea here of putting online multiple > > mail servers working off single NFS mounted partition. > > It's about 9Gb+ of /var/mail shared between 3-4 hosts > > ( we'll probably do round-robin DNS for them or will > > just have MX with same priorities to redistribute > > the load evenly). > > Is your current mailserver overloaded? Have you been able to identify > the bottleneck(s)? Is it really compute power that's the limitation? > I'm kinda skeptical here, unless you are receiving _lots_ of mail. > Going for a multinode arrangement will increase your compute power and > (possibly) your network bandwidth, but it won't help the setup/knockdown > time for connections and it will decrease your disk bandwidth. > > > So the Big Question is: will it WORK ??? > > Er. You might have to fiddle sendmail to lock mailboxes in an alternative > fashion. > > > Will it be fast enough ( the thing should serve ~20.000 - 50.000 > > active e-mail accounts), will there be problems with locking > > mail-boxes for delivery and reading , what's gonna happen > > to the network , it there a sense in using 100Mb Ethernet, > > how stable is NFS code ? > > NFS locking isn't yet supported. Your network will potentially be > plastered with lots of NFS traffic in addition to the mail. > > What counts as 'active'? If we assume that 5 mails a day is 'average', > at 50 000 users that's about three a second. I'd say that's well within > the capacity of a medium-sized mailserver (but I'd want to test it to be > sure 8) > > If you get this thing running, make sure you put something up on the Web > about it; a mailserver that size must be a pretty rare beast. Personally, I am somewhat skeptical that there isn't a better way to do this. NFS is traditionally just about the worst way to do mail, IMHO. However, a more detailed explanation of the environment and desired result would be needed. > > If the thing will work, may be we will do the same with news > > system. > > For news, you want one machine hosting the news, and a number of NFS > clients running readers only. Don't try to run your news databases > across NFS, it's just too slow. NFS clients running readers? I disagree..! You are not lowering the I/O demands on the server disks at all. Run slaves :-) > > Does FreeBD support any of them RAID arrays ? Or the only way is > > to get one in H/W implementation ? > > The ccd driver may well be up to running news - Joe Greco? ccd works fine, I have not tested more than two disks striped, more than one ccd device, or a partition > 8GB, but I do have 2-drive 4GB and 8GB partitions running on news.sol.net and daily-planet.execpc.com, very reliable. news.sol.net has actually STOPPED crashing because it's no longer running out of disk space every few days :-) ... Joe ------------------------------------------------------------------------------- Joe Greco - Systems Administrator jgreco@ns.sol.net Solaria Public Access UNIX - Milwaukee, WI 414/546-7968