Performance of Top Link with clustering

Question

Hi,
We are planning to use Top Link with a J2EE app that will reside on Oracle 9i app server. We are also planning to use clustering of app server. So we will need to configure Top Link ServerSessions accordingly (isnt it?).
Do you think the performance of TopLink will be reduced badly if used in this fashion, since the server sessions have to talk to each other continuously !
Thanks,
Krishna

Answer 1

Krishna,
The Cache-Sync feature of TopLink is an excellent way to minimize stale data. The messaging and change-set propagation does not come without a cost though. Typically we try to recommend this for customers with a high percentage of reads. In these cases it is more beneficial to keep the cache and avoid trips to the database.
There is no magic formula for when to use cache-sync. The messaging cost/efficiency is dependent on the mechanism (point to point over RMI/CORBA/IIOP or JMS), the size of the change sets or complexity of transaction, number of servers in the cluster, etc.
What is important to understand though is that cache-sync is just one way to minimize stale cache. It does not ensure you won't have stale data. For this you must configure a locking strategy and adjust your cache relative to the expected life-span you need on cached objects.
For a system like your I would recommend starting out without cache-sync. Use optimistic locking on all classes and use a weak cache for all classes that have a high occurrence of writes. For the rest I would recommend a Full cache on static/reference data and a soft-weak cache on those that fall in between. The size of the soft-weak cache will determine the number of objects potentially held (LRU) beyond their use.
With these settings you should be able to monitor and account for changes made to stale objects through OptimisticLockExceptions. If you find specific use-cases that are prone to failure you could force the refresh on the query at the start of the UnitOfWork (TX).
This approach will give you a system that ensures data integrity. Then you can experiment with scaling to multiple servers with or without cache-sync and monitor the load vs failures for your test environment.
I hope this helps,
Doug