> If you share your assertions or whatever, there will be a cost in CPU and
> network traffic to do that synchronisation

You're correct that there would be a network latency cost for any
conceivable solution, but I'm fairly certain network latency per se is
largely negligible in a capable HA setup.  The cost of synchronization
would vary with the storage backend:

 - JBossCache:  notable, but smaller these days with buddy replication
 - Terracotta: likely considerable, but don't have enough experience to estimate
 - memcached: zero
 - RDBMS: zero

> Depending on the kind of synchronization, you can also consume really more
> memory : instead of having 1000 assertions on each node (for n nodes),
> you will have n * 1000 assertions on each node (each node has all assertions).

I would argue N-way synchronization is a poor solution for
non-critical authenticated state data.  If you have concerns of
exceeding storage limits, buddy replication would be a better
solution.

> At limit, the synchronization cost will exceed the gain to add a node in 
> cluster

I don't have any numbers to challenge that assertion, but I'm fairly
confident you've overstated it quite a bit with the technologies we
would consider as storage backends for the Java client.

M

-- 
You are currently subscribed to [email protected] as: 
[email protected]
To unsubscribe, change settings or access archives, see 
http://www.ja-sig.org/wiki/display/JSG/cas-dev

Reply via email to