Hazelcast, Terracotta and IMDGs

For years I've been wanting Terracotta to come up with a comprehensive and polished IMDG (In Memory DataGrid) solution using their fantastic technology. It seems like such an obvious use case, but it's never really happened. Obviously, you can distribute a map with Terracotta but a real IMDG needs much more. You can achieve it all with TC but it requires a lot of manual work. The closest thing they have right now is tim-searchable and it still not the full stack. I headed about a product some time ago called Hazelcast but when I looked at it, it was REALLY rough around the edges, but I took another look last night and it seems to have come along quite nicely. It has full write-through and write-behind capabilities and the grid is fully queryable. I was always a little leery of their pure partitioning scheme which seemed to ignore the benefits of data locality to each instance, but they do now have a “near cache” concept that can help alleviate that issue to some degree. From a technology standpoint, it's light years behind Terracotta and I would presume that translates to lower performance, but there are thresholds of performance that are more than sufficient for different apps and it seems like Hazelcast may be well above the threshold that many apps would find acceptable. I’m going to check it out sometime in the next week or two.

Comments

[JonDo]

Hi,
did you have more new thoughts about hazelcast/terracotta? i'm now in the middle of choosing which one to use, the performance of hazelcast looks very good compared to memcached (around 80% of the memcached performance) and i didn't had time to check terracotta performance yet, but for now it looks like hazelcast is a very good solution and much more scalable (on paper) than terracotta...

[Matthew Welch]

Sorry. I never really got beyond the evaluation phase. The next time I need IMDG capabilities I'll be spending some more time with it to see if it would meet my needs.