From owner-freebsd-hackers Wed Feb 2 10:12:44 2000 Delivered-To: freebsd-hackers@freebsd.org Received: from acl.lanl.gov (acl.lanl.gov [128.165.147.1]) by builder.freebsd.org (Postfix) with ESMTP id ADD4E4017 for ; Wed, 2 Feb 2000 10:12:41 -0800 (PST) Received: from mini.acl.lanl.gov (root@mini.acl.lanl.gov [128.165.147.34]) by acl.lanl.gov (8.8.8/8.8.5) with ESMTP id LAA4075333 for ; Wed, 2 Feb 2000 11:07:47 -0700 (MST) Received: from localhost (rminnich@localhost) by mini.acl.lanl.gov (8.9.3/8.8.8) with ESMTP id LAA01345 for ; Wed, 2 Feb 2000 11:07:47 -0700 X-Authentication-Warning: mini.acl.lanl.gov: rminnich owned process doing -bs Date: Wed, 2 Feb 2000 11:07:46 -0700 (MST) From: "Ronald G. Minnich" X-Sender: rminnich@mini.acl.lanl.gov To: freebsd-hackers@FreeBSD.ORG Subject: Re: Re/Fwd: freebsd specific search In-Reply-To: Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: owner-freebsd-hackers@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.ORG On Wed, 2 Feb 2000, Michael Bacarella wrote: > Granted, a lot of Linux distributions are totally unsuited for a server > environment. Compared to that, I could understand why the > server-orientedness of FreeBSD is attractive, but I certainly couldn't put > up a reasonable arguement for either side in Slackware Linux vs. FreeBSD. Linux is definitely a less reliable system for clustering than freebsd. I've got 5 years of running them both at Sarnoff to back me up. Maybe I was doing something wrong, but I'm seeing similar problems here at the ACL on Linux. We ran into four classes of problems that linux had that freebsd did not. These problems are still not fixed as of 2.2.x or 2.3.x. 1) network stack. heavy use of udp can result in a hung kernel. Trivial TCP servers that need to take lots of connections cause trouble -- clients start getting ECONNREFUSED after a while 2) nfs. Hit nfs hard and random clients will hang. The dirty little secret of linux clusters is that 'everyone knows' that you don't run client nfs on linux cluster nodes if you want the cluster to stay up. This came out clearly at a cluster conference last spring (JCP4). 3) vm system. There's still some strange problems in there. 4) ext2. ext2 does not handle unplanned outages well. There is a reasonable chance that after a power fail you're going to have trouble if you have 100 nodes or more. You'll see 2 or 3 in need of help. freebsd was just more solid on our clusters. But note that linux isn't standing still -- it's just not as good as freebsd yet. I had one freebsd cluster that ran through 5 years of anything we could throw at it -- power fail, etc. It took disk death to finally halt one node and require me to hook up a keyboard to it to reload it. Our general experience was that NT fails a lot, esp. if you ask it to talk to a network or run a screensaver. Linux clusters run a long time, but power outages and other unplanned events will cause it trouble. Freebsd tolerates very high levels of abuse. The UFS guys really know their stuff. ron To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hackers" in the body of the message