Does using WKA instead of multicast have negative side effects?

Question

Will it impact Performance?

Answer 1

Hi Colin,
I would not expect the performance benefits of multicast to help you out here unless the query (not the result) is very very large, i.e. multi megabyte. Even when unicast is used, we still send the requests out to all nodes asynchronously, so the increase in latency over the multicast approach would be minimal.
If you are seeing query times increase as the cluster grows it would be more likely attributable to lack of indexs on your caches, or not having indexes on the attributes used in your query. Without an index Coherence is forced to not only iterate every cached item, but also deserialize them (yes it is stored in serialized form).
Finally cluster wide queries are by their vary nature not scalable unless you've also increased the query set size. Think of it this way, if you have N records in M nodes and perform a query each node will do the following work:
Q + N/M + A
Where Q means process the question, and A means send an answer. If N increases and M decreases each node will be able to compute the answer quicker (and it will trend towards zero). But each node still must perform the constants Q and A, so from a cluster wide perspective you are performing M*(Q+A) work and M is increasing. Now if N is also increasing you will still see scalability benefits, as it would have taken a smaller cluster longer to run the query on the bigger data set, but for a fixed size data set you should see diminishing returns.
In such cases it would be useful to have a look at KeyAssociation, an KeyAssociatedFilter. This approach allows for scalable queries even at fixed size data sets.
http://download.oracle.com/otn_hosted_doc/coherence/342/com/tangosol/net/cache/KeyAssociation.html
http://download.oracle.com/otn_hosted_doc/coherence/342/com/tangosol/util/filter/KeyAssociatedFilter.html
thanks,
Mark
Coherence Development Team