3 Replies Latest reply on Apr 4, 2017 3:16 AM by Radim Vansa

    Infinispan 9 vs Chronicle

    Hristo Stoyanov Master

      Hi all,

      I am trying to select Infinispan or Java Chronicle for some high-performance budding product. Has anyone compared the perfirmance of both and willing to share information?


      My observation is that the people behind Java Chronicle claim significantly better perfirmance, especially since their product us used for high-frequency trading.



        • 1. Re: Infinispan 9 vs Chronicle
          Radim Vansa Master

          What exactly would you like to compare? Infinispan is a data-grid, with ConcurrentMap-like API and goodies like indexing (search by non-primary keys), events etc., and Chronicle is a persistent queue with JMS-like APIs. If you want to push notifications from one node to another, use Chronicle; if you want to store data and eventually read them, use Infinispan. Note that while we consider allocation an important factor in Infinispan (as it affects performance), ATM we don't try to generate zero garbage during steady state.

          • 2. Re: Infinispan 9 vs Chronicle
            Hristo Stoyanov Master

            Thanks Radim,

            As I mentioned this is for a high-performance bidding application where response time is critical. Both products offer Map-like structures with primitive search, events, network protocols. Both can use off-heap storage (to avoid GC) and both claim to use fast binary serialization. One would expect about the same performance, but Chronicle is used in the high-frequency-trading community (HFT), which may imply that its throughput of events is way better than the usual Java datagrids (Infinispan, Hazelcast, Apache Ignite).  I am note sure if this is the case, hence my performance comparison question.  Infinispan makes things a lot easier as being part of Wildfly/Swarm, but I would not sacrifice response time performance just for that.

            • 3. Re: Infinispan 9 vs Chronicle
              Radim Vansa Master

              Can't help you here, then. If you want to run some benchmarks, I suggest implementing OpenHFT plugin in RadarGun , a tool we use a lot for performance comparisons. Note, however, that RG itself is not zero-garbage and therefore it could throw away some benefits that OpenHFT collections bring.

              Anyway, if you need any online help with that, contact us on IRC.