0 Replies Latest reply on May 21, 2009 9:47 AM by alexey.ragozin

    Using custom backing map for distributed scheme


      We are evaluating usage of distributed scheme for storing large amount of data. We have analyzed memory usage for distributed-scheme backed by local-scheme (with binary unit calculator) and received interesting results.

      On single JVM
      110M - byte arrays - our data in binary form
      20M - com.tangosol.util.Binary objects - overhead of binary objects to wrap byte arrays
      27M - com.tangosol.net.cache.LocalCache$Entry objects - local cache entry overhead

      When we have tried our own implementation of java.util.Map as backing map (using some clever tricks to get smaller size at the cost of CPU processing)

      On single JVM
      36M - byte arrays - out application data

      So it is look like we are going to stick with our implementation.

      My question
      Are there any known drawbacks of using implementation of java.util.Map as backing map for distribute-scheme?
      Our implementation is thread safe of cause (we have already tested it thread-count option of distributed-scheme)