RE: [JBoss-user] If http session is replicated.. no need for sessionaffinity?
> I agree. Plus you get better performance: with sticky sessions, you only > serialize once per session modification, otherwise, you pay the price twice. Sorry, but I don't see what you mean. You're saying that replica servers do NOT need to deserealize the sessions they get, unless they need to use them? Otherwise, I see no logic in your statement of serialize 1 time vs serialize 2 times. What I think is that: 1 - Cluster nodes A,B,C 2 - Session1 is created in A - A serializes Session1 and sends to group by multicast - B and C get the message, and unserialize session1. They now have session1 replica in memory 3a)- Session1 is changed in A (sticky session load balancing) - A serializes Session1 and sends to group by multicast - B and C get the message, and unserialize session1. They now have session1 replica in memory 3b)- Session1 is changed in B (LB without sticky session) - B serializes Session1 and sends to group by multicast - A and C get the message, and unserialize session1. They now have session1 replica in memory I see no performance difference in using either a dunmb LB or a sticky session LB. So, if there is, the cenario I just described is wrong. Please describe me why! Thank you Joao Clemente --- This sf.net email is sponsored by: Jabber - The world's fastest growing real-time communications platform! Don't just IM. Build it in! http://www.jabber.com/osdn/xim ___ JBoss-user mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/jboss-user
Re: [JBoss-user] If http session is replicated.. no need for sessionaffinity?
> If we know that your cluster is using session affinity, there are a > large number of optimisations that we can make, which will result in a > faster service for you. > I will be checking in an updated implementation of distributable > sessions for Jetty shortly, which I shall then spend some time > optimising in this way. Ok. But that is a feature to come? There is no optimization at this time? I mean, there is no advantage at this time? > You would be well advised to use session affinity if you can. If you > don't you are simply making a lot more work for your cluster. Well, I am trying to understand WHY I am doing more work for my cluster this way. Can you explain that? Lets supose the replication occurs without problems, the requests are spaced in time no there are no problems in incoerent states. Is there a penalty (memory usage, performace, or whatever) for not directing the requests to the same instance? Why? --- This sf.net email is sponsored by: Jabber - The world's fastest growing real-time communications platform! Don't just IM. Build it in! http://www.jabber.com/osdn/xim ___ JBoss-user mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/jboss-user
Re: [JBoss-user] If http session is replicated.. no need for sessionaffinity?
If we know that your cluster is using session affinity, there are a large number of optimisations that we can make, which will result in a faster service for you. I will be checking in an updated implementation of distributable sessions for Jetty shortly, which I shall then spend some time optimising in this way. You would be well advised to use session affinity if you can. If you don't you are simply making a lot more work for your cluster. Jules Joao Clemente wrote: > Hi. > As I was looking for a good load balancing solution for JBoss, it just came to my >mind this question: > > If httpSession is replicated, all machines have all session instances in memory... >So, there is no need to use a "sticky-session" load balancing algorithm, right? > Furthermore, as we do not need to check the session, we don't need a layer-7 LB, so >we can use a simpler/faster layer-4 LB without a problem! > > Or maybe there is something I am not seeing and it's still better to redirect the >client to the same machine? > --- This sf.net email is sponsored by: Jabber - The world's fastest growing real-time communications platform! Don't just IM. Build it in! http://www.jabber.com/osdn/xim ___ JBoss-user mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/jboss-user
Re: [JBoss-user] If http session is replicated.. no need for sessionaffinity?
If we know that your cluster is using session affinity, there are a large number of optimisations that we can make, which will result in a faster service for you. I will be checking in an updated implementation of distributable sessions for Jetty shortly, which I shall then spend some time optimising in this way. You would be well advised to use session affinity if you can. If you don't you are simply making a lot more work for your cluster. Jules Joao Clemente wrote: > Hi. > As I was looking for a good load balancing solution for JBoss, it just came to my >mind this question: > > If httpSession is replicated, all machines have all session instances in memory... >So, there is no need to use a "sticky-session" load balancing algorithm, right? > Furthermore, as we do not need to check the session, we don't need a layer-7 LB, so >we can use a simpler/faster layer-4 LB without a problem! > > Or maybe there is something I am not seeing and it's still better to redirect the >client to the same machine? > --- This sf.net email is sponsored by: Jabber - The world's fastest growing real-time communications platform! Don't just IM. Build it in! http://www.jabber.com/osdn/xim ___ JBoss-user mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/jboss-user