Re: [jgroups-dev] Question about JGroups and Coherence

I copied the JGroups dev mailing list (pls send mails to it next time, 
instead of to me directly). Comments below.

Kuwon Kang wrote:
> Hello Bela.
> I'm kuwon from South Korea.
>
> Few months ago I read the article about *
> ReplCache.*
> *And I found that ReplCache implemented almost of distributed cache 
> system based on JGroups.*
> *
> *
> *Our company adapted Oracle Coherence for massive data to reduce total 
> time spending for processing data.*
> *
> *
> *As I know that one of Coherence's power thing is communication 
> performance
> between nodes. And their max hopes are two. *

ReplCache uses concistent hashing, so the max hop is always 1.

> *Anyway If you compare the performance for network communication between
> Coherence and JGroups in clusters.*

I haven't. ReplCache was a prototype that I wrote to experiment with a 
distributed cache; the idea was then used to implement the distributed 
cache (DIST mode) in Infinispan.

So if you want to compare 2 distributed cache, you need to compare 
Coherence and Infinispan, *not* Coherence and JGroups. JGroups' focus is 
a reliable transport, not a cache.

> What's the differences between two for network performance

I suggest write your own performance test, which mimics the load and 
access patterns of your application. There's also the Cache Benchmark 
Framework [1], which you could use to compare different caches, e.g. 
Infinispan and Coherence.


> If there are big performance differences, is it good idea to adapt 
> JGroups for mission critical system for processing massive data using 
> distributed
> cache nodes like using ReplCache.

ReplCache is only a prototype, so I would not use it in mission critical 
environments. However, it's relatively small and based on the (very 
mature and battle tested) JGroups, so I don't think you'll run into too 
many issues... but again, my recommendation is Infinispan rather than 
ReplCache.

> In the case of our project, Coherence reduced 30% time decrement using 
> 24 nodes.

The time to process data was reduced by 30% using 24 rather than 1 node 
? That would be marginal, should be much more ! But I probably misread 
your question.

> I think any other factors will not be problem for performance but only one
> thing that communication performance will be the one.
>
> The business application retrieves so many times to the database, this
> causes some performance problem. So that's the reason to use distributed
> cache system like Coherence.

Or Infinispan, give it a try !


[1] https://sourceforge.net/projects/cachebenchfwk/

-- 
Bela Ban
Lead JGroups / Clustering Team
JBoss

------------------------------------------------------------------------------
This SF.net email is sponsored by 

Make an app they can't live without
Enter the BlackBerry Developer Challenge
http://p.sf.net/sfu/RIM-dev2dev 
_______________________________________________
Javagroups-development mailing list