[prev in list] [next in list] [prev in thread] [next in thread] 

List:       jboss-user
Subject:    Re: [JBoss-user] basic clustering question
From:       "Brian McSweeney" <brian.mcsweeney () aurium ! net>
Date:       2003-01-31 11:06:02
[Download RAW message or body]

thanks Jon,


> The "router" is a proxy running on the client side, and each time you
> communicate with one of your beans, the client  execute code that
> decides which server to ask.  Every time the cluster discovers that
> the cluster configuration changes, the client proxy is updated
> (whenever communication is done).

this makes sense if the client is a heavyweight app (say swing). But
if the client is the web tier and the app is basically an internet site,
then
the ultimate client is a web browser. Now, you could have the "proxy"
being the code in the web tier - ie, a servlet, which could pick which of
the ejb tiers to hit, as you say.

However I would have thought it would be better to let this proxy be a
little hardware device that just selects which of the boxes to hit, and then
each box could run both the web and ejb tiers and make each web tier
distributable. What do yo think?
Brian


----- Original Message -----
From: "Jon Haugsand" <Jon-H.Haugsand@norges-bank.no>
To: <jboss-user@lists.sourceforge.net>
Sent: Friday, January 31, 2003 8:02 AM
Subject: Re: [JBoss-user] basic clustering question


>
> (Sacha is the right man here, but I'll try anyway.)
>
> * Brian McSweeney
> >
> > 1) I assume that we would have a simple router infront of each of the
> > boxes
> > to handle load balancing and just picking one of the boxes at random to
> > send
> > requests to. Is this how it's normally done?
>
> The "router" is a proxy running on the client side, and each time you
> communicate with one of your beans, the client  execute code that
> decides which server to ask.  Every time the cluster discovers that
> the cluster configuration changes, the client proxy is updated
> (whenever communication is done).
>
> >
> > 2) do they both have to use one instance of a database eg on box 3, if
> > so,
> > isn't this a single point of failure anyway? How is this normally done?
>
> This has bothered me as well.  Another problem is that we use
> messaging (actually IBM's MQ Series) and there is also a single point
> of failure.
>
> >
> > 3) in order to cluster, do all the beans have to be remote to keep the
> > other
> > nodes updated, or can the entity beans be local?
>
> Entity beans are updated through the database.  There is some caching
> and locking issues though, and I think Sacha et al is working on some
> distributed cache and/or lock.  The interesting case is the stateful
> session beans.
>
> --
>  Jon Haugsand, Jon-H.Haugsand@norges-bank.no
>  http://www.norges-bank.no
>
>
>
> -------------------------------------------------------
> This SF.NET email is sponsored by:
> SourceForge Enterprise Edition + IBM + LinuxWorld = Something 2 See!
> http://www.vasoftware.com
> _______________________________________________
> JBoss-user mailing list
> JBoss-user@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/jboss-user
>




-------------------------------------------------------
This SF.NET email is sponsored by:
SourceForge Enterprise Edition + IBM + LinuxWorld = Something 2 See!
http://www.vasoftware.com
_______________________________________________
JBoss-user mailing list
JBoss-user@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/jboss-user
[prev in list] [next in list] [prev in thread] [next in thread]