在Java集群环境中管理会话数据的有效方法是什么？

Question

suppose we have an internet web site with high visits, say, 10M pv per day,

say,

1. session sticky can not handle failover in a nice way, the users will be impacted;
2. session replication may cause too much overhead,
3. use some clustered cache server for storing actual session data and only do session replication for things like key (say, useid)?

I once asked someone in ebay and they say they use in memory mysql clustering for this. any good ways/best practice?

Answer 1

In memory cluster is best option, see

http://www.mysql.com/products/cluster/features.html

Answer 2

You can try using Terracotta Websessions

• http://terracotta.org/
• http://terracotta.org/products/web-sessions

Answer 3

well, since it's surely not mean distribution, i do not have the data yet, but i'd expect it should at least handle 1000 dynamic request per second if possible

It is possible to have bursts of this much activity, which is why I asked. The solution is different depending on what you believe the burst to be. You can support 1K updates per second, with MySQL provided you have a fast disk (disk controller cached or SSD)

BTW: I have a library for a database which supports 200K - 20M updates per second on commodity hardware, but its not as easy to use, so you really need to need it. ;)