[cas-dev] high availability issue of CAS

[David Harrison]

On 19/11/2008, at 10:46 PM, Lin George wrote:

> 1. From your reply below, your recommended way is to set replication  
> information between clustered CAS servers so that all servers  
> contain all user session information, i.e. all cache of all  
> clustered CAS servers contain idential information? Then when one  
> server is down, the other server could be used as backup?
>
> (For example, if a new user login from one CAS server, the CAS  
> server stores user credential/session information in its cache, and  
> then replicate the information into another CAS server cache?)

The JBoss Cluster documentation is very handy when it comes to setting  
up a CAS cluster:
http://docs.jboss.org/jbossas/jboss4guide/r4/html/cluster.chapt.html

Alternatively if JBoss is not your thing the clustering capabilities  
of Glassfish v2 are very good (and much easier to use):
http://docs.sun.com/app/docs/doc/819-3193/gatqf?a=view

Note: In both of these cases you need to setup some form of ticket  
replication as described in the CAS clustering wiki.


> 2. I am not sure how many user data is able to be stored into one  
> server, do you have any information? The target user size is about  
> 200k users (including users which are graduated) for my school.  
> Thanks.

This is difficult to answer without knowing the usage patterns of your  
users, the session expiry time and what other applications (if any)  
will be running concurrently.

The general rule is the more RAM the better - and don't forget to  
configure JBoss to use it.
e.g. Set appropriate -Xms and -Xmx JAVA_OPTS in JBoss's bin/run.conf  
file


A sensible approach would be to setup a basic cluster and use some  
form of monitoring to record the user and memory load over time.

For this I like to use Hyperic as it has some really nice JBoss/Tomcat  
monitoring features:
http://www.hyperic.com/
However there are many simpler and more complex things out there for  
doing this sort of thing.


You may find a two member CAS cluster is perfectly happy - or it may  
just implode under the load.
The suspense is what makes this sort of thing fun :-)


David






>
>
> regards,
> George
>
> --------------------
> Most of our high-availability clustering shares information across
> servers such that if one goes down the other can handle whatever the
> downed server was doing.
>
> -Scott
>
> -Scott Battaglia
> PGP Public Key Id: 0x383733AA
> LinkedIn: http://www.linkedin.com/in/scottbattaglia
>
>
>
>
> _______________________________________________
> cas-dev mailing list
> cas-dev at tp.its.yale.edu
> http://tp.its.yale.edu/mailman/listinfo/cas-dev

David Harrison
http://www.linkedin.com/in/dhharrison