Kevin Martin
1995-Nov-10 10:21 UTC
Anyone else think it's about time to beat a WEB server to death?
> What I'd most ideally like would be some numbers from a site that's to > WEB servers what ftp.cdrom.com is to FTP servers, but I'll take whatever > I can get! :-)> Anyone got any stats they'd like to share? # of running daemons, > server used, hits-per-second, hardware used, that kind of thing.Check these excerpts of daily posted statistics from BEST Internet (www.best.com). They get enormous traffic, too much for their two Pentium FreeBSD systems (I think they want to do use SGI soon, but they'll screw it up). I'm not sure who you would contact for detailed information about their configuration. I think they're P5-120's with 128Mb RAM, perhaps? From: root@best.net (Root) Newsgroups: best.status Subject: Daily WWW HIT Usage Report for Best Date: 10 Nov 1995 02:53:54 -0800 Organization: Best Internet Communications +TOTAL+ 2643688 21479.201 MB doki 266846 1359.395 MB iminfo 175801 1104.376 MB trnelson 102914 1034.096 MB asixinc 54528 682.670 MB webmastr 12226 541.634 MB walterh 22783 454.667 MB stw 30886 377.983 MB haynes 56291 371.855 MB etzine 72022 358.696 MB dml 14246 346.098 MB craig 30897 310.661 MB eroticsf 31107 278.579 MB single 22695 273.475 MB onethumb 2599 264.150 MB rusty 18844 260.804 MB backdrop 30349 235.669 MB 9latigid 29594 225.976 MB icc 2899 222.078 MB raptor 10451 219.839 MB From: root@best.net (Root) Newsgroups: best.status Subject: Daily WWW CGI Usage Report for Best Date: 10 Nov 1995 02:54:20 -0800 x /home/ftpadmin/root/pub/haynes/public_html 6130 0.000 MB x /home/httpd/pub/imagine/public_html 5921 0.000 MB x /home/httpd/pub/robj/public_html 4042 0.000 MB x /home/ftpadmin/root/pub/blowfish/public_html 3718 0.000 MB x /home/ftpadmin/root/pub/entcafe/public_html 3387 0.000 MB x /home/httpd/pub/9latigid/public_html 3335 0.000 MB x /home/ftpadmin/root/pub/vincek/public_html 2557 0.000 MB x /home/ftpadmin/root/pub/myee/public_html 2541 0.000 MB x /home/httpd/pub/vance/public_html 2442 0.000 MB x /home/ftpadmin/root/pub/robj/unreal/public_html 1990 0.000 MB x /home/ftpadmin/root/pub/cean/public_html 1842 0.000 MB x /home/ftpadmin/root/pub/romantsy/public_html 1748 0.000 MB x /home/ftpadmin/root/pub/bizcafe/public_html 1455 0.000 MB x /home/ftpadmin/root/pub/edhall/public_html 1419 0.000 MB x /home/httpd/pub/mccanna/public_html 1417 0.000 MB x /home/ftpadmin/root/pub/cbntmkr/public_html 1331 0.000 MB x /home/ftpadmin/root/pub/artemis/fatfree/public_html 1207 0.000 MB x /home/ftpadmin/root/pub/agraps/public_html 1179 0.000 MB x /home/ftpadmin/root/pub/gobills/public_html 1134 0.000 MB x /home/ftpadmin/root/pub/malch/public_html 1127 0.000 MB x /home/ftpadmin/root/pub/ecat/public_html 1104 0.000 MB x /home/ftpadmin/root/pub/faith/public_html 1072 0.000 MB x /home/ftpadmin/root/pub/crunch/public_html 1063 0.000 MB x /home/httpd/pub/jh/public_html 1060 0.000 MB x /home/ftpadmin/root/pub/raul/public_html 1000 0.000 MB
Rashid Karimov
1995-Nov-10 11:03 UTC
Anyone else think it's about time to beat a WEB server to death?
Hi there folx,> > I frequently get asked the question: "How many users can I run off a > FreeBSD WEB server?" and I'm naturally tempted to ask in response > "How long is a piece of string?" > > However, I check myself with the knowledge that it's not an entirely > unreasonable thing to want to know, and I merely wish that I had more > data on this subject to provide in response. It's obviously > impossible to come up with one number that fits all situations, but > various guesstimates can be derived from existing data so that given a > link speed of x, a PC of macho-factor y and the "average" user doing > z, you can come up with a performance projection of n users. > > The only problem is that I don't *have* any existing data worth > mentioning. > > WEB servers what ftp.cdrom.com is to FTP servers, but I'll take whatever > I can get! :-)OK , here is some quite impressive stuff: I have here a Web Server ( NCSA 142, slightly patched by me to work with multiple domains ... wasn't able to find the native implementation) which runs about 50 different Web sites now ( on Ip aliases ) , including famous www.jumbo.com. The former gets about ~200.000 hits a day .. lemme see exactly how many hits I have basing on access_log file ( it is exactly 1 day old now): 409059 strings ! It is 2.1.Stable on P-90 PCI/SCSI ( Adaptec 2940) with 64 Mb of RAM. I have 100 httpd processes spawned at the start-up time and 200 max - after that httpd will behave in the old fashioned way, spawing a child per request. The beast in incredibly fast - at least comparing to old NCSA httpd 1.3 . It serves about 2 Gb of html/text files The other data I have on FreeBSD as a server is: I run FreeBSD based PCs as the shell/POP/ftp servers here. Average is about 4000-5000 account per PC. There are usually 300+ processes running on the system at the peak time , about 40-60 users logged in. Load averages range from 0.2 to 2.5 Uptimes are around 50 days - usually I have to reboot server because of some kind of maintenance/HW upgrade before it dies on its own :). I also have a few PCs as news ( INND) and DNS servers. That is completely bullet proof - servers stay up forever :)> Heck, if you need some additional incentive for signing up for such a > mad scheme then might I suggest also putting up some adverts for > whatever service you offer on the page as "live data" (grin) so all > those hundreds (thousands?) of users will also see your advertising in > the process of trying to see how much punishment a FreeBSD WEB server > can take.. We could even make it more widely publicised challenge by > posting details of the event in various non-FreeBSD newsgroups, like > Linux's or BSDI's. Given an open invite to see if they can bring a > FreeBSD WEB server to its knees, I'm sure many of the "competing OS" > advocates wouldn't be able to resist a challenge like that, especially > if the testing authority promised in advance to be relatively > impartial and post full results, be they good or bad. I'm confident > enough in this product that I think we'd come out looking pretty good! > > Either way, it would also generate a lot of publicity for all > concerned (us and the test machine providers) and furnish the FreeBSD > Project with some very valuable data that it doesn't have now.Yes, I think it's possible to use www.jumbo.com with adv. purposes - and everybody is welcome to test how fast it is :) Rashid
Terry Lambert
1995-Nov-10 11:07 UTC
Anyone else think it's about time to beat a WEB server to death?
> I frequently get asked the question: "How many users can I run off a > FreeBSD WEB server?" and I'm naturally tempted to ask in response > "How long is a piece of string?" > > However, I check myself with the knowledge that it's not an entirely > unreasonable thing to want to know, and I merely wish that I had more > data on this subject to provide in response. It's obviously > impossible to come up with one number that fits all situations, but > various guesstimates can be derived from existing data so that given a > link speed of x, a PC of macho-factor y and the "average" user doing > z, you can come up with a performance projection of n users. > > The only problem is that I don't *have* any existing data worth > mentioning.If the httpd is started from inetd, then the limit is dictated by no more than 256 requests in any 60 second period, unless you override this at inetd startup time by increasing the number of requests allowed per 60 seconds using a -R when you start it in the rc file. If you don't override this, your test will fail in short order with "server failing (looping), service terminated". You can kill most BSD inetd based FTP servers this way now, actually, using a -d 0 on an ncftp retry when the server is already loaded. I saw ftp.mv.com go down the other day when an associate stupidly kicked the retry delay to 0. By default, it takes 10 minutes for the thing to reset and start serving connection requests again (#define RETRYTIME (60*10) in inetd.c). I first saw this problem with a machine tftp serving fonts, color tables, and configuration files to a very large number of X terminals back in the 1.x days, and posted the equivalent of the '-R' option addition at that time. Terry Lambert terry@lambert.org --- Any opinions in this posting are my own and not those of my present or previous employers.
Michael Smith
1995-Nov-10 13:28 UTC
Anyone else think it's about time to beat a WEB server to death?
Jordan K. Hubbard stands accused of saying:> However, I check myself with the knowledge that it's not an entirely > unreasonable thing to want to know, and I merely wish that I had more > data on this subject to provide in response. It's obviously > impossible to come up with one number that fits all situations, but > various guesstimates can be derived from existing data so that given a > link speed of x, a PC of macho-factor y and the "average" user doing > z, you can come up with a performance projection of n users. > > The only problem is that I don't *have* any existing data worth > mentioning.Brian Tao's benchmarks (700K+ hits per day) are a good opener when talking to commercial vendors that think that 10K/hour is "heavy". A little while ago we had someone here with a problem where they were peaking at 100/sec but falling off to 70/sec until update ran. I haven't heard any more since then on that one. I realise that neither of these are 'real-world' cases.> If we can't get any actual data from existing WEB service providers, > or even if we can, might I prevail on someone out there with a > well-connected box to possibly declare a "flag day", during which as > many people on this list as possible (and anyone else they can find) > aggressively attempts to beat the server to its knees while the server > maintainers busily collect stats on the event?Do a 'dry run' first 8) -- ]] Mike Smith, Software Engineer msmith@atrad.adelaide.edu.au [[ ]] Genesis Software genesis@atrad.adelaide.edu.au [[ ]] High-speed data acquisition and (GSM mobile) 041-122-496 [[ ]] realtime instrument control (ph/fax) +61-8-267-3039 [[ ]] My car has "demand start" -Terry Lambert UNIX: live FreeBSD or die! [[
Jordan K. Hubbard
1995-Nov-10 16:51 UTC
Anyone else think it's about time to beat a WEB server to death?
I frequently get asked the question: "How many users can I run off a FreeBSD WEB server?" and I'm naturally tempted to ask in response "How long is a piece of string?" However, I check myself with the knowledge that it's not an entirely unreasonable thing to want to know, and I merely wish that I had more data on this subject to provide in response. It's obviously impossible to come up with one number that fits all situations, but various guesstimates can be derived from existing data so that given a link speed of x, a PC of macho-factor y and the "average" user doing z, you can come up with a performance projection of n users. The only problem is that I don't *have* any existing data worth mentioning. I've really only two WEB servers that I can honestly say I have much experience with, and that's www.cdrom.com and www.freebsd.org. Unfortunately, www.cdrom.com gets so little WEB traffic in comparison to FTP traffic (on which we have LOTS of data) that the numbers are almost lost in the noise. www.freebsd.org is also a popular server as servers go, but not so popular that we're getting dozens of hits per second or anything as impressive sounding as that. What I'd most ideally like would be some numbers from a site that's to WEB servers what ftp.cdrom.com is to FTP servers, but I'll take whatever I can get! :-) Anyone got any stats they'd like to share? # of running daemons, server used, hits-per-second, hardware used, that kind of thing. If we can't get any actual data from existing WEB service providers, or even if we can, might I prevail on someone out there with a well-connected box to possibly declare a "flag day", during which as many people on this list as possible (and anyone else they can find) aggressively attempts to beat the server to its knees while the server maintainers busily collect stats on the event? Heck, if you need some additional incentive for signing up for such a mad scheme then might I suggest also putting up some adverts for whatever service you offer on the page as "live data" (grin) so all those hundreds (thousands?) of users will also see your advertising in the process of trying to see how much punishment a FreeBSD WEB server can take.. We could even make it more widely publicised challenge by posting details of the event in various non-FreeBSD newsgroups, like Linux's or BSDI's. Given an open invite to see if they can bring a FreeBSD WEB server to its knees, I'm sure many of the "competing OS" advocates wouldn't be able to resist a challenge like that, especially if the testing authority promised in advance to be relatively impartial and post full results, be they good or bad. I'm confident enough in this product that I think we'd come out looking pretty good! Either way, it would also generate a lot of publicity for all concerned (us and the test machine providers) and furnish the FreeBSD Project with some very valuable data that it doesn't have now. So how about it? Any takers? If you're really interested in helping to further the cause of Spreading The Word, I can assure you that this would be a significant step in the right direction. I'll also be more than happy to work with whomever steps forward in drafting a reasonably provocative-sounding announcement to ensure that people take up the gauntlet. After all, how much trouble can whapping "reload" for 5 or so minutes be? :-) Jordan P.S. Suggestions on how to make this an even more meaningful test from those webmaniacs out there among you would also be sincerely appreciated!>From root@fyeung5 Sat Nov 11 00:44:38 1995Status: RO X-Status: Received: from blob.best.net (blob.best.net [204.156.128.88]) by shellx.best.com (950911.SGI.8.6.12.PATCH825/8.6.5) with ESMTP id AAA06989 for <netific@shellx.best.com>; Sat, 11 Nov 1995 00:44:37 GMT Received: from bsdi.BSDI.COM (bsdi.BSDI.COM [205.230.224.1]) by blob.best.net (8.6.12/8.6.5) with ESMTP id QAA22220 for <fyeung@netific.com>; Fri, 10 Nov 1995 16:44:41 -0800 Received: from austin.bsdi.com (daemon#r4MJMfstq8MnmHM5cGDHZGcdJZn827pH#@austin.BSDI.COM [205.230.229.1]) by bsdi.BSDI.COM (8.6.12/8.6.12) with ESMTP id OAA03794 for <inet-access-dist@bsdi.com>; Fri, 10 Nov 1995 14:58:51 -0700 Received: (from daemon@localhost) by austin.bsdi.com (8.6.12/8.6.12) id PAA15968 for inet-access-dist@earth.com; Fri, 10 Nov 1995 15:58:49 -0600 Resent-Date: Fri, 10 Nov 1995 15:58:49 -0600 Resent-Message-Id: <199511102158.PAA15968@austin.bsdi.com> List-Admin: inet-access-request@earth.com (subscribe/unsubscribe requests) Errors-To: owner-inet-access@earth.com Originator: inet-access@earth.com Precedence: bulk Reply-To: inet-access@earth.com Resent-From: inet-access@earth.com Sender: inet-access@earth.com Received: from caboose.ironhorse.com (root@sl-caboose.ironhorse.com [204.145.167.1]) by austin.bsdi.com (8.6.12/8.6.12) with ESMTP id PAA15964 for <inet-access@earth.com>; Fri, 10 Nov 1995 15:58:45 -0600 Received: from caboose.ironhorse.com (petesk@caboose.ironhorse.com [204.145.167.2]) by caboose.ironhorse.com (8.6.12/8.6.9) with SMTP id NAA18708 for <inet-access@earth.com>; Fri, 10 Nov 1995 13:59:30 -0800 Date: Fri, 10 Nov 1995 13:59:29 -0800 (PST) From: "Peter J. Skelly" <petesk@ironhorse.com> To: inet-access@earth.com Subject: Re: PC w/ Linux In-Reply-To: <Pine.SUN.3.91.951110142824.16625H-100000@millenium.texas.net> Message-ID: <Pine.LNX.3.91.951110135256.18663A-100000@caboose.ironhorse.com> MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII On Fri, 10 Nov 1995, Jonah Barron Yokubaitis wrote:> On Thu, 9 Nov 1995, Peter J. Skelly wrote: > > (Obviously, our customer bases, and therefore usage patterns differ. > > Thats why I'm able to get away with the news system I have, and probably > > why you need the news system you have) > > laugh, go run 200 news clients on your 16M server. Your machine will swap > so heavily you won't be able to telnet to it. > > Compare apples to apples.Why do you think I included the parenthetical statement... I am fully aware that our userbases are different. In this whole discussion, I do not think ANY of the Linux newsadmins implied that they were expecting to run 200, or even as many as 20, readers on their news machines. Why is it that people running large ISP's can't seem to see that those of us running smaller ISP's may not want 3000 whining, non net-aware, support intensive, 10$ a month users. That is not the only business model in existence. Peter J. Skelly petesk@ironhorse.com Ironhorse Software, Inc. (206) 999-9983 Windows (NT), OLE, LINUX, OS/2 Software Development http://www.ironhorse.com/~petesk
Satoshi Asami
1995-Nov-10 18:28 UTC
Anyone else think it's about time to beat a WEB server to death?
Just FYI, I deleted the looping address (fyeung@fyeung5.netific.com) from the mailing list file manually. (I didn't go through majordomo because it would generate an "majordomo results" message which would be fed back to this list again.) Can someone add this address to the "rejected" list? I don't know how to do it.... The Gunslinger