forums.smartfoxserver.com

I've looked over all of the forum posts that I can find on this subject. Given the size of our client and their insistence on what the client load will likely look like at certain times of year it's been decided that we do indeed need a cluster of servers.

Now that being decided, we've been rather hard pressed to find any information as to how to accomplish this with SmartFoxServer 2X. The clustering documentation for the previous version of the SmartFoxServer, while full of great information, references the manipulation of many files that just don't seem to exist in this build.

Any chance someone could point me in the right direction?

I sent an email to info [AT] smartfoxserver earlier today, but figured I should reach out to the community as well. This is an issue we'll need to be on the road to solving rather quickly.

Thanks.
Dan

Hi,
I will report here my email reply:

Our documentation for clustering SFS with Terracotta refers to the 2.x version of the Terracotta SDK. Unfortunately with the introduction of version 3.x (over a year ago) the tool has steered from its original intention and has become a very sophisticated caching system.
They have hidden all of the DSO from the developer's perspective and they don't even recommend to use it, which is too bad.
This way Terracotta doesn't work anymore as a transparent clustering engine, instead it requires a tight integration with their API as any other similar technology. This is why we discontinued the integration with SFS2X

what about running sfs on a beowolf cluster. im not entirely sure what a linux beowolf operating system would allow you. but in theory if it did you could add new nodes when needed with minimal work.

Beowulf cluster the OS, not the applications.
The simplest and most effective way to cluster SFS is to run as many instances you need in the cloud (for example using RightScale which we support directly)

The key question is "How to share state" across multiple servers?
Answer --> It depends.
Most of our customers do it via a central DB cluster. Why? Because a DB cluster is easier to setup than a socket-based server and because there are many options on which solution to use (RDBMS like MySQL, NOSQL such as Couch or MongoDB etc...)