From owner-freebsd-isp Fri Sep 20 17:18:52 2002 Delivered-To: freebsd-isp@freebsd.org Received: from mx1.FreeBSD.org (mx1.FreeBSD.org [216.136.204.125]) by hub.freebsd.org (Postfix) with ESMTP id D0D6E37B401 for ; Fri, 20 Sep 2002 17:18:50 -0700 (PDT) Received: from loops.nilpotent.org (loops.nilpotent.org [12.17.163.70]) by mx1.FreeBSD.org (Postfix) with SMTP id 1D28943E65 for ; Fri, 20 Sep 2002 17:18:50 -0700 (PDT) (envelope-from fn@hungry.org) Received: (qmail 29890 invoked from network); 20 Sep 2002 20:18:43 -0000 Received: from unknown (root@203.215.177.158) by 12.17.163.70 with QMTP; 20 Sep 2002 20:18:43 -0000 Received: (qmail 61212 invoked by uid 500); 20 Sep 2002 20:12:06 -0000 To: freebsd-isp@freebsd.org Cc: VGirnet@megadat.com Subject: Re: WCCP performance questions X-nil: X-Useless-info: System load is 1.06 with 63 processes active. X-Neuromancer: What's in Beme, okay? X-Linux-Version: Cappy Ishmael's Moby Linux From: fn@hungry.org (Faried Nawaz) Organization: Integral Domains Date: Sat, 21 Sep 2002 02:12:06 +0600 Message-ID: Lines: 35 User-Agent: Gnus/5.090008 (Oort Gnus v0.08) XEmacs/21.4 (Artificial Intelligence, i386-unknown-freebsd4.6) MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Sender: owner-freebsd-isp@FreeBSD.ORG Precedence: bulk List-ID: List-Archive: (Web Archive) List-Help: (List Instructions) List-Subscribe: List-Unsubscribe: X-Loop: FreeBSD.org At an ISP I do random work for, I have three Pentium III 1GHz boxes set up with 512MB RAM and 12GB of dedicated cache space on IDE drives each. They're running FreeBSD 4.6-STABLE (updated today to 4.7-RC) with squid 2.4.6 using diskd. They're doing WCCP off a Cisco 3640 (hint: make sure you use "ip cef"). They used to handle 10-20 requests/sec, but the ISP's expanded recently and they've hit 35+ requests/sec. At peak hours there are 600 dialup users online. From the MRTG graphs, it doesn't seem like they the proxies are maxed out yet -- they're still responding to requests in very short timeframes. All three proxies show 50-75% of data being served out of cache. I might experiment with using heap LFUDA as the cache replacement policy to see if it helps for my type of users. cache_mem for squid is set to 170MB (~ 1/3rd of main memory). Each proxy is set up with the other two as cache peers, and each proxy also runs its own caching nameserver (dnscache, from djbdns) which uses up 40MB of RAM. FreeBSD on the systems uses ipfilter to route packets to squid. The only special kernel config options I used were to set maxusers to 0 and to use 8192 mbuf clusters (more may or may not be needed for your mix of users; use "netstat -m" to guide you). squid isn't very CPU-intensive (uses ~ 20-30% CPU on my boxes). I don't see how multiple processors will help, since it's not multithreaded. However, maxing out on memory and disk space will definitely be useful. Faried. -- The Great GNU has arrived, infidels, behold his wrath ! If I wanted a GF, Values, not variables. I'd use Dylan. To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-isp" in the body of the message