From owner-freebsd-chat Sat Oct 24 22:27:32 1998 Return-Path: Received: (from majordom@localhost) by hub.freebsd.org (8.8.8/8.8.8) id WAA08457 for freebsd-chat-outgoing; Sat, 24 Oct 1998 22:27:32 -0700 (PDT) (envelope-from owner-freebsd-chat@FreeBSD.ORG) Received: from bang.rain.com (bang.rain.com [204.119.8.73]) by hub.freebsd.org (8.8.8/8.8.8) with SMTP id WAA08451 for ; Sat, 24 Oct 1998 22:27:29 -0700 (PDT) (envelope-from john@bang.rain.com) Received: (from john@localhost) by bang.rain.com (8.6.12/8.6.9) id WAA17958; Sat, 24 Oct 1998 22:26:53 -0700 From: John Cavanaugh Message-Id: <199810250526.WAA17958@bang.rain.com> Subject: Re: another record To: dg@root.com Date: Sat, 24 Oct 1998 22:26:52 -0700 (PDT) Cc: chat@FreeBSD.ORG In-Reply-To: <199810241823.LAA05442@implode.root.com> from "David Greenman" at Oct 24, 98 11:23:44 am X-Mailer: ELM [version 2.4 PL24] Content-Type: text Sender: owner-freebsd-chat@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.org > >> Well, yesterday's killer record of 587GB from wcarchive didn't last > >> long. It's getting a bit "interesting" that we're so close to topping the > >> terabyte/day threshold. > > > >Why "interesting"? > > > >We still have a ways to go before you saturate the 100Mb ethernet don't we? > >(don't skewer me if i'm wrong, I haven't done the math) And I know that > >the Pro/100B isn't the "sticky" point as far as pumping data out to the > >rest of the world... > > The fast ethernet was maxed out for most of the day. It will be necessary > to increase our circuit bandwidth before we'll be able to go much higher than > this. Average packet size is less than 1000 bytes. Layer 2 packet overhead > limits us to around 85-90Mbps with full duplex fast ether. The addition of > layer 3 overhead reduces the actual throughput by even more. There is also > more data going out than just files being downloaded (welcome message, > messages that come out when you cd to various directories, directory listings, > etc. - none of this is accounted for in the stats)...this actually amounts to > more than you might think. The totals we're talking about only include the > total number of downloaded file bytes sent out. Wow. So we are somewhat topped out then. It still amazes me that this is just one box. ;-) > >Can you give us some more details on wcarchives other upcoming upgrades? > > > >You had mentioned putting a 400Mhz Xeon in. Is this change going to a > >allow more ftp sessions or just get the load average under 30 occasionally? > > > > > >Thanks! ;-) > > The load average on a machine like wcarchive might just as well be a random > number. It's a composite of both disk and CPU "load" and isn't useful in our > case for determining the machine's potential. In fact, what is interesting > is that the load average (which is typically around 25-60 on wcarchive, but > varies a lot) is so LOW. Don't forget that we have *3500* file downloads go > on. One might expect the load average to be well into the hundreds. Oh yeah, I wasn't meaning I wasn't impressed - just looking for what the benefits of the Xeon were going to be. > As for planned upgrades, we'll be going to Xeon/4xx in a month or so. The > main reason for doing this is the increased memory capacity - the new machine > will have 4GB of RAM. This will allow us to increase the FTP limit to at > least 10000 users. My main concern at the moment is that we don't have > sufficient network bandwidth to support that many users (we're just hitting > the limit of our 100Mbps circuit with 3500 users). We're talking with CRL > about our options. I'm advocating gigabit ethernet, but we may have to > settle for multiple 100Mbps circuits in the short term. Either way - that will be even more impressive. So, if we went specifically with the math from above, those 10000 users would be able to fill up about 3 100Mbps circuts. Would the Xeon be able to fill all of those pipes? (I guess you wouldn't be considering it if it wouldn't now would you? ;-)) Is the plan to just keep growing wcarchive to meet demand or maybe split the duties across multiple machines? Looking at the stats that you've posted recently, it looks like the FreeBSD directories always get a lot of traffic, but when some new first person shooter (or in this case, 3rd person shooter - Heretic II), comes out, those directories traffic goes through the roof. Do you think it would be better to carve off certain sections of the main archive for bandwidth reasons? What solution do you think is the best? Thanks. -- John Cavanaugh "There can be only one." To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-chat" in the body of the message