<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN"> <html xmlns:v="urn:schemas-microsoft-com:vml" xmlns:o="urn:schemas-microsoft-com:office:office" xmlns:w="urn:schemas-microsoft-com:office:word" xmlns:st1="urn:schemas-microsoft-com:office:smarttags" xmlns="http://www.w3.org/TR/REC-html40"> <head> <meta http-equiv=Content-Type content="text/html; charset=us-ascii"> <meta name=ProgId content=Word.Document> <meta name=Generator content="Microsoft Word 10"> <meta name=Originator content="Microsoft Word 10"> <link rel=File-List href="cid:filelist.xml@01C199EB.06469DA0"> <title>Scale of server</title> <o:SmartTagType namespaceuri="urn:schemas-microsoft-com:office:smarttags" name="time"/> <o:SmartTagType namespaceuri="urn:schemas-microsoft-com:office:smarttags" name="date"/> <o:SmartTagType namespaceuri="urn:schemas-microsoft-com:office:smarttags" name="PersonName"/>  <style>  </style>  </head> <body bgcolor=white lang=EN-US link=blue vlink=purple style='tab-interval:.5in'> <div class=Section1> I’m actually testing load on a server right now.  I have one box with three instances of dpsm, one instance of jabberd, as well as the yahoo and msn transports running.  The user information is stored in a mySQL db running on the same machine and jabber accesses it through a version of xdb_sql that we have modified to fix a few quirks and to make it work with the msn transport.  The way we are load balancing the connections to the three instances of dpsm is to just run each one on a virtual ip and set up our dns to round robin the three ip’s.<o:p></o:p> <o:p> </o:p> With a straight test that does nothing but connect sockets to the server, we can get 10,000 connections per instance of dpsm.  Three instances of dpsm seems to be the limit with this test, as there are diminishing returns with any more than that.  I’m now using a tester app I found in the CVS repository to get an actual load test.  The app has some problems with it so we are tweaking it and running tests.  I don’t have solid numbers yet, but I know that I have had 3,000 users simultaneously logged in and all of them sending messages.  The problem with this test is that if you have 3,000 people logged in at any given time, they will not be constantly talking to each other.  So with some tweaking we are trying to make a load test that will more accurately simulate a load.<o:p></o:p> <o:p> </o:p> What we are trying to accomplish here is to have 30,000 clients connected to the one box, with a percentage of them talking that would accurately reflect a real life server.<o:p></o:p> <o:p> </o:p> Note:  Our server machine is just a desktop that we stole to test on.  It’s a one CPU P4 1.3Ghz I believe, with 1GB RAM.  If we had multiple CPU’s I suspect we could get away with running even more instances of dpsm.  We are also going to try adding 2 more network cards to the machine so each instance of dpsm will be on it’s own network card.  Let me know if you need any more info.  I’ve found the information on scaling Jabber to be pretty lean and hard to find.<o:p></o:p> <o:p> </o:p> Jason Reineri<o:p></o:p> <st1:PersonName>Northridge</st1:PersonName> Systems, Inc.<o:p></o:p> <o:p> </o:p> -----Original Message----- From: Michael F. March [mailto:march@indirect.com] Sent: <st1:date Month="1" Day="10" Year="2002">Thursday, January 10, 2002</st1:date> <st1:time Hour="14" Minute="38">2:38 PM</st1:time> To: jdev@jabber.org Subject: Re: [JDEV] Scale of server <o:p> </o:p> <div> Do you know of a Jabber server that has had 3K users on at one time?<o:p></o:p> </div> <div>  <o:p></o:p> </div> <blockquote style='border:none;border-left:solid black 1.5pt;padding:0in 0in 0in 4.0pt; margin-left:3.75pt;margin-top:5.0pt;margin-right:0in;margin-bottom:5.0pt'> <div> ----- Original Message ----- <o:p></o:p> </div> <div style='font-color:black'> From: <a href="mailto:jreineri@northridge.com" title="jreineri@northridge.com">Jason Reineri</a> <o:p></o:p> </div> <div> To: <a href="mailto:jdev@jabber.org" title="jdev@jabber.org">jdev@jabber.org</a> <o:p></o:p> </div> <div> Sent: <st1:date Month="1" Day="10" Year="2002">Thursday, January 10, 2002</st1:date> <st1:time Hour="7" Minute="29">7:29 AM</st1:time><o:p></o:p> </div> <div> Subject: RE: [JDEV] Scale of server<o:p></o:p> </div> <div> <o:p> </o:p> </div> Jabber can handle 3000 concurrent users with no problem.  The only problem you will have is that the select() function in Linux is hard coded to a certain number of sockets.  In kernel 2.4 that limit is 1024.  You can either recompile the kernel to allow more, or you can use dpsm or jpolld as your socket manager.  These socket managers use the poll() function instead of select() and are not subject to the limit.  You will however have to increase the number of file descriptors you can have open per process.  This can be done in a startup script that calls ulimit.<o:p></o:p> <o:p> </o:p> Jason Reineri<o:p></o:p> <st1:PersonName>Northridge</st1:PersonName> Systems, Inc.<o:p></o:p> <o:p> </o:p> -----Original Message----- From: Rodrigo Roman [mailto:dedalo@ferengi.com.ar] Sent: <st1:date Month="1" Day="9" Year="2002">Wednesday, January 09, 2002</st1:date> <st1:time Hour="18" Minute="14">6:14 PM</st1:time> To: 'jdev@jabber.org' Subject: [JDEV] Scale of server<o:p></o:p> <o:p> </o:p> I'm planing an instalation of jabber server in a isp, we have aprox 3000 concurrent users (i dont know wich porcentage will use our messanger) our client base is 150k users.<o:p></o:p> Wich is the best implementation for this numbers? Is jabber prepared for this? Are all the transports able to work in this client base? <o:p></o:p> Thanks for your help! <o:p></o:p> </blockquote> </div> </body> </html>