I played with clustering ODE a bit. I think the problem with clustering is that it doesn't have common public API implemented by various vendors. Instead of that, each vendor has it's own clustering extensions. For example, Axis2 uses Tomcat's tribes. So I think we need to support various cluster implementations in ODE in future. This will include SMX3 & 4 too. So let's make an interface, which will include cluster related operations, like noticing new node or rewriting http destination while making outgoing http calls from ODE. This could have pluggable implementations for various clustering vendors.
On 7 April 2010 04:40, Jeff Yu <[email protected]> wrote: > Hi all, > > I'd like to bring this topic up, > https://issues.apache.org/jira/browse/ODE-563, I am going to work on this > task, before I do this, I'd like to have a heads up at here. > > Below is discussion that I found in the mail archive by Alex Boisvert, which > I think it is very good summary in this area. > > " > The Axis2 integration already supports *clustering*. (Well, maybe a better > characterization is that it doesn't get in the way) > > There's a few things do to to support *clustering*. First thing would be to > > make the scheduler *cluster*-aware, to support work distribution (load > balancing) across different nodes and to avoid contention. I suggested > using Terracotta as a basis for this, but there may be other (and better) > ways. > > If we want to get more performance out of the system, we can introduce > varying levels of node affinity to reduce data shipping and improve cache > efficiency within the *cluster*. This would probably entail some form of > distributed lock manager, or a registry to redirect messages to process > instances already loaded on *cluster* nodes. This is an area where > integration with the messaging layer (Axis2) can help since you can > establish "sessions" between clients and services to improve message routing > > and, again, cache efficiency. > > The persistence DAOs (OpenJPA and Hibernate) also support *clustering*today. > Both can be configured to be *cluster*-aware and both support the more > aggressive transparent *cluster* cache such as Terracotta, JBoss Cache, etc. > > > Hope this helps, > alex > " > > From above summary, I believe we have following sub-tasks for the > clustering. > > 1. make cronScheduler cluster-aware. > 2. simpleScheduler module's DAO need to be implemented as JPA based, so that > it can support cluster. > 3. make simple scheduler cluster-aware? > > Anything I am missing or mis-understanding here?? > > -- > Cheers, > Jeff Yu > > ---------------- > blog: http://jeff.familyyu.net > Regards, -- Rafał Rusin http://rrusin.blogspot.com http://www.touk.pl http://top.touk.pl
