From owner-freebsd-stable Wed Jan 31 16:13:38 2001 Delivered-To: freebsd-stable@freebsd.org Received: from utterlux.hq.communitconnect.com (unknown [206.28.215.90]) by hub.freebsd.org (Postfix) with ESMTP id 9C35C37B4EC; Wed, 31 Jan 2001 16:13:15 -0800 (PST) Received: from localhost (marius@localhost) by utterlux.hq.communitconnect.com (8.11.1/8.11.1) with ESMTP id f110D9l52099; Wed, 31 Jan 2001 19:13:14 -0500 (EST) (envelope-from marius@mail.communityconnect.com) X-Authentication-Warning: utterlux.hq.communitconnect.com: marius owned process doing -bs Date: Wed, 31 Jan 2001 19:13:09 -0500 (EST) From: Marius X-Sender: marius@utterlux.hq.communitconnect.com To: stable@FreeBSD.org Cc: questions@FreeBSD.org Subject: Multiple dropped thttpd connections under high load. Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: owner-freebsd-stable@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.ORG My company has a number of FreeBSD 3.5-Stable servers running thttpd 2.2.17, they happily buzz along under very high load. (With a 'options NMBCLUSTERS=8192' in the kernel config to give us some serious mbufs.) So a few months back I went ahead and upgraded one of the servers to 4.x-Stable. It currently is running 4.1.1-Stable. (But I am upgrading it now to 4.2-Stable as I write. Some of those recent security advisories struck a chord.) Anyway, since upgrading from 3.5-Stable to 4.1.1.-Stable the performance of the machine and it's webserver has been flakey whe under higher load. Under calmer conditions it works fine, but when the traffic raises we get a high occurance of dropped connections. Even just telneting to the webserver from the local host can create a dropped conection. (Sometimes as it is contecting the webserver, and sometimes while it is trying to fulfill the requests.) The 3.5-Stable machines don't exhibit this behavior. The problem of dropped connections has become so common that I have been told to take it out of service. But I really want to get it back into service. Someday thttpd will have support for 'accept filters' and I want to be ready. I could -so- use them. I am running the same thttpd binary as I did on 3.5-Stable. I copied over a fresh image from a companion machine recently. Since it works under light/medium load on the machine, I didn't think it needed to be recompiled. mbuff (netstat -m) usage can be high, but I don't think I am running out of them. I've never seen it at more the 75%. Does anybody have some suggestions for improving/tuning things to work better? I know this situation is rather specific, if everybody was having problems like this, there would be PR's on file, but I can't find any. And I can't remember any chatter like this in recent months on Stable. Any suggestions/educated-guesses would be appreciated. -Marius M. Rex To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-stable" in the body of the message