From owner-freebsd-questions Fri Nov 17 16:33:54 1995 Return-Path: owner-questions Received: (from root@localhost) by freefall.freebsd.org (8.6.12/8.6.6) id QAA09754 for questions-outgoing; Fri, 17 Nov 1995 16:33:54 -0800 Received: from mistery.mcafee.com (jimd@mistery.mcafee.com [192.187.128.69]) by freefall.freebsd.org (8.6.12/8.6.6) with ESMTP id QAA09742 for ; Fri, 17 Nov 1995 16:33:43 -0800 Received: (from jimd@localhost) by mistery.mcafee.com (8.6.11/8.6.9) id RAA07661 for questions@freebsd.org; Fri, 17 Nov 1995 17:35:55 -0800 From: Jim Dennis Message-Id: <199511180135.RAA07661@mistery.mcafee.com> Subject: MAXUSERS=64 -- kernel panics To: questions@freebsd.org Date: Fri, 17 Nov 1995 17:35:55 -0800 (PST) X-Mailer: ELM [version 2.4 PL24] MIME-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Content-Length: 2370 Sender: owner-questions@freebsd.org Precedence: bulk Hey, I recently installed FreeBSD 2.0.5 to replace my overloaded Linux box. We run an ftp site that almost as popular as Walnut Creek's (roughly 400,000 file transfers per month). Naturally I found that the generic kernel would run out of file handles almost immediately (5 min. max.) that was at 24 incoming sessions. Besides I certainly planned on rebuilding the kernel anyway. So I config'd to 128 -- it complained that this was > 64, so I brought it down to 64 and rebuilt. Stop reading if you've heard this one before..... Ahh -- this must be news then -- she reboots! The 2.0 kernel (which I had tried first) would wait ten minutes (getting upto 70 ftp sessions or so) and then panic. 2.0.5 had the grace to refrain from lulling me into a false sense of security by dying immediately. It was a simple matter to reboot on the generic kernel, 'ifconfig de0 down' and build one with MAXUSERS set to 32. That was stable but would run out of file handles at ~95+ users (under a half hour). So I built her again with a setting of 63 and brought her back up. I've been monitoring her for a week now with a cron job to mail me ftpcount's every hour and an entry in rc.local to mail me a dmesg on every reboot (root is .forwarded to another box inside -- from which I'm mailing this). There have been no more reboots. The ftpcount has been averaging 150 and the I wrote a test to start 200 concurrent processes (while 50 users were in from outside) at which she didn't even blink. So, moral of that story is -- don't set MAXUSERS to 64. This is a Pentium 90 with 64Mb of RAM, NCR PCI SCSI and a DEC PCI ethernet (only about a Gig online). Obviously I don't know if more RAM would allow me to get past this point. I suspect that MAXUSERS=63 will allow this machine to keep our T1 (and the new second one that's on order) completely flooded for some time to come. Since I put this up I'm not getting any more complaints about being unable to connect. I'm getting some about transfer rates -- but as I said -- we've already got another T1 on order. The Sun3 is still serving 100 concurrent (almost all day and night) so we're running an aggregate of 250 concurrent connections on average. The webserver is also getting about 100,000 times a day. Jim Dennis, System Administrator, McAfee Associates