users@shoal.java.net

Re: [Shoal-Users] dsc scalability question

From: Shreedhar Ganapathy <Shreedhar.Ganapathy_at_Sun.COM>
Date: Sun, 20 Apr 2008 21:55:35 -0700

HI leehui

leehui wrote:
> Hi,
>
> Recently, I use DistributedStateCache (gms.getGroupHandle().getDistributedStateCache()) to store some data.
> The cache is global. I mean, if one group member changes the cache's value, the value must replicate to all the
> other group members. Is that true?
>
That is true.
> If so, when you put a lot of objects into cache, the overload of network will be expensive. How should I do when I
> store lots of objects via DistributedStateCache?
>
What is the size of each object and how many per second are you
expecting to save into the cache ? How many threads will be writing data
and how many would be reading and at what rate?
This data will be helpful to characterize performance.

Our goal was to provide a global cache for lightweight throughput. A
massively scalable cache is something we are looking to provide in the
near future.

The current distributed cache needs optimizations as a global cache in
order to support large number of objects.
> Another question is when shoal uses p2p commnunication,and when shoal uses group communication. Look at following api (GroupHandle's interface).
>
> (1)void sendMessage(String targetComponentName, byte[] message) throws GMSException;
> (2)void sendMessage(String targetServerToken, String targetComponentName,byte[] message) throws GMSException;
> (3)void sendMessage(List<String> targetServerTokens, String targetComponentName,byte[] message) throws GMSException;
>
> Can I say that (1) uses group communication while (2)&(3) use p2p communication?
>
Currently we use p2p for all messaging including group level messaging
as UDP is unreliable.

The advantage with the p2p approach is that Jxta provides NIO based
channels in that case. This allows for high scaling capabilities.

A light weight reliable multicast class exists in the provider
implementation but is not extensively tested from a performance perspective.

>
> Thank you in advance for your reply!
>
>

> --------------
> leehui
> 2008-04-21
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: users-unsubscribe_at_shoal.dev.java.net
> For additional commands, e-mail: users-help_at_shoal.dev.java.net
>
>