From owner-freebsd-questions Fri May 5 20:41:18 1995 Return-Path: questions-owner Received: (from majordom@localhost) by freefall.cdrom.com (8.6.10/8.6.6) id UAA22152 for questions-outgoing; Fri, 5 May 1995 20:41:18 -0700 Received: from aries.ibms.sinica.edu.tw ([140.109.40.248]) by freefall.cdrom.com (8.6.10/8.6.6) with ESMTP id UAA22146 for ; Fri, 5 May 1995 20:41:10 -0700 Received: (from taob@localhost) by aries.ibms.sinica.edu.tw (8.6.11/8.6.9) id LAA07812; Sat, 6 May 1995 11:37:23 +0800 Date: Sat, 6 May 1995 11:37:22 +0800 (CST) From: Brian Tao To: David Greenman cc: FREEBSD-QUESTIONS-L Subject: Re: Heavy HTTPD serving on 2.0-950412 In-Reply-To: <199505051840.LAA04219@corbin.Root.COM> Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: questions-owner@FreeBSD.org Precedence: bulk On Fri, 5 May 1995, David Greenman wrote: > > options "NMBCLUSTERS=1024" Yep, found it in the mail archives after John Fieber updated them on the Web site. > This will give you twice as many buffers (or 4 times if you don't have > GATEWAY in your kernel). Double the above to 2048 if the problem > persists. I use 8192 on wcarchive, which is enough for 1000 ftp users > (about 2000 TCP connections). Ah, my next two questions already answered. ;-) I ran an 11-hour test with NCSA httpd 1.4 last night with 4 local Ethernet clients and 60 remote clients. No problems with NMBCLUSTERS set to 1024. Doing a 'netstat -n | grep ^tcp' showed between 450 and 520 open connections, almost all in TIME_WAIT and nearly 16K in the send queue (not surprisingly, all to sites across the Pacific). I haven't done the analysis yet, but it looks like NCSA's preforking helps out a whole lot here. I'm guessing about 600,000 accesses per day with a load average of less than 1.00. :) 'systat -mbuf' shows this: /0 /1 /2 /3 /4 /5 /6 /7 /8 /9 /10 Load Average | /0 /5 /10 /15 /20 /25 /30 /35 /40 /45 /50 /55 /60 pcbs XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX 94 socknames XXXXXXXXXXXXXXXXXXX data XXXXXX headers XXXXX Ummm... what should I be looking for, if I think I'm running out of mbufs? Add up all the X's and see if they exceed NMBCLUSTERS? How about checking a kernel's NMBCLUSTERS setting? I didn't see anything in the sysctl man page. > >Active Internet connections > >Proto Recv-Q Send-Q Local Address Foreign Address (state) > >netstat: kvm_read: kvm_read: Bad address > > Rebuild netstat; the networking structurs in the kernel have changed. Got it. Thanks.