what i always wonder.. How do you sync sessions? That's why i think none sticky sessions are just flawed..
Do remember that we in wicket have a single threaded mode. How on earth are you going to do that if you have none sticky sessions? If request a comes in alters the page, but then at +/- at the same time another request comes in to another server that also changes the page... How does the server ever merge that? The last one to leave wins? But that could be the first one that did enter. And if those 2 request would have altered 2 different parts of the page? (ajax requests) What do we have then in memory? I just can't believe that the servlet container can really merge the objects And even then what happens when there is a conflict? So the only thing to do here for a servlet container is to synchronize over the servers But then the whole point of none sticky sessions is just gone. The overhead would be huge. Besides that what does a none sticky session bring you if you talk about performance? do you really thing it is then better balanced? I don't think so. If the numbers are high then because of probability everything will average out for example if you start at 0 then round robin you divide new sessions to 4 different servers.. after 2000 new sessions the servers have 500 sessions each.. then 400 sessions are removed the probabililty tells us that each server will have about +/- 400 sessions left. and not 1 server is dropped to 100 and the other 3 do still have 500 sessions. johan On Nov 16, 2007 6:19 PM, Igor Vaynberg <[EMAIL PROTECTED]> wrote: > round-robin has a couple of disadvantages over sticky sessons: > > a) you need to replicate state to all nodes - a lot of traffic > although with a fiber backplane its most likely not an issue. with > sticky sessions you only have to replicate to one or two other nodes > which act as backup buddies. > > b) you cannot take advantage of data locality. each node can keep its > own cache which over a span of a session gets more attuned to what the > users on that node are doing, if you use round robin then you lose out > on that. > > -igor > > > On Nov 16, 2007 8:51 AM, saenz <[EMAIL PROTECTED]> wrote: > > > > Fair enough. Good answer. :) > > > > Of course I meant in the non-failure case. > > > > So under normal operations outside of failure conditions, perhaps there > > really is no benefit to move users to other nodes during the lifetime of > a > > session, or at least the difficulty outweighs the benefit. > > > > - Lu > > > > > > > > > > Martijn Dashorst wrote: > > > > > > failover? > > > > > > On 11/16/07, saenz <[EMAIL PROTECTED]> wrote: > > >> > > >> > > >> Johan Compagner wrote: > > >> > > > >> > On Nov 15, 2007 11:22 PM, Eelco Hillenius < > [EMAIL PROTECTED]> > > >> > wrote: > > >> >> Recommended is to use sticky sessions. > > >> > > > >> > +10000 > > >> > > > >> > none sticky sessions are flawed anyway.. (or the synchronizing > > >> performance > > >> > hit is so great it doesn't make any sense, it it makes sense to > start > > >> > with...) > > >> > > > >> > > >> Johan, are you saying that other types of load balancing strategies > > >> besides > > >> sticky sessions are all flawed, and that it only makes sense to use > > >> sticky > > >> sessions? Please correct my understanding. > > >> > > >> If sticky sessions are the only recommended way to do load balancing, > > >> then > > >> what is the reason to support replication of session state across > > >> multiple > > >> nodes in a cluster? > > >> > > >> - Lu > > >> -- > > >> View this message in context: > > >> > http://www.nabble.com/Disabling-serialization-storage-of-pages-in-session--tf4768006.html#a13796491 > > >> Sent from the Wicket - User mailing list archive at > > >> Nabble.com<http://nabble.com/> > . > > >> > > >> > > >> --------------------------------------------------------------------- > > >> To unsubscribe, e-mail: [EMAIL PROTECTED] > > >> For additional commands, e-mail: [EMAIL PROTECTED] > > >> > > >> > > > > > > > > > -- > > > Buy Wicket in Action: http://manning.com/dashorst > > > Apache Wicket 1.3.0-rc1 is released > > > Get it now: http://www.apache.org/dyn/closer.cgi/wicket/1.3.0-rc1/ > > > > > > --------------------------------------------------------------------- > > > To unsubscribe, e-mail: [EMAIL PROTECTED] > > > For additional commands, e-mail: [EMAIL PROTECTED] > > > > > > > > > > > > > -- > > View this message in context: > http://www.nabble.com/Disabling-serialization-storage-of-pages-in-session--tf4768006.html#a13796995 > > > > Sent from the Wicket - User mailing list archive at > > Nabble.com<http://nabble.com/> > . > > > > > > --------------------------------------------------------------------- > > To unsubscribe, e-mail: [EMAIL PROTECTED] > > For additional commands, e-mail: [EMAIL PROTECTED] > > > > > > --------------------------------------------------------------------- > To unsubscribe, e-mail: [EMAIL PROTECTED] > For additional commands, e-mail: [EMAIL PROTECTED] > >
