Hi Steve, I came across the SCTP protocol the other day (http://en.wikipedia.org/wiki/Stream_Control_Transmission_Protocol <http://en.wikipedia.org/wiki/Stream_Control_Transmission_Protocol+>) and it mentioned that there was a protocol that is built on top of it called Reliable Server Pooling *(*http://en.wikipedia.org/wiki/Reliable_server_pooling and http://tdrwww.iem.uni-due.de/dreibholz/rserpool/ ). Reading up on both of them got me to thinking about how a distributed, partitioned cache, or Terracotta as well, could sit on top of these protocols to enable a peer-to-peer computing grid.
SCTP has open source implementations for Linux and several flavors of UNIX, as well as an implementation for Windows, and it looks like Reliable Server Pooling has an open source implementation for UNIX as well. I haven't downloaded them and opened them up, but I think it could be worth investigating for use in WADI and Terracotta. Although there isn't a Java version written yet, I'm sure it could be done and would be a neat by-product of Terracotta development. If I recall, Terracotta is aiming to go to a primary-to-primary architecture eventually, and this could be one approach to enabling it. Cheers, Jim On 7/5/07, Steven Harris <[EMAIL PROTECTED] > wrote:
Thx filip, I'll look at that. Cheers, Steve On Jul 5, 2007, at 12:31 PM, Filip Hanik - Dev Lists wrote: Steven Harris wrote: I've been thinking about what the best way in the terracotta world is to deal with cache eviction. If someone is building an app on top of terracotta and needs eviction one might start with an LRUMap. In a distributed world this has the problem of lock contention and even has limits in the single vm case depending on how time critical your cache is. I naturally started investigating Clock caches (nice treatment over in the derby world) http://wiki.apache.org/db-derby/DerbyLruCacheManager But in a distributed world this doesn't really completely solve the problem either. It would be expensive to replicate the changes to the items in the cache to all nodes and the clock cache doesn't seem great for a virtual heap because it seems it would fault in the world. Seems like a user needs a partition-able concurrent cache eviction policy. Has anyone done work in this area? One who did a lot of work (and now is passive/retired from the project) is Jules Gosnell on WADI. Basically that was a session replication mechanism, but deal a great deal with partitions and how to move partitions around in the cluster. http://wadi.codehaus.org/ if nothing else, you can touch base with him, and he might have some good pointers, not to WADI, but to ideas, since WADI is specialized Filip Cheers, Steve ------------------------------------------------------------------------ _______________________________________________ tc-dev mailing list [email protected] http://lists.terracotta.org/mailman/listinfo/tc-dev ------------------------------------------------------------------------ No virus found in this incoming message. Checked by AVG Free Edition. Version: 7.5.476 / Virus Database: 269.9.14/885 - Release Date: 7/3/2007 10:02 AM _______________________________________________ tc-dev mailing list [email protected] http://lists.terracotta.org/mailman/listinfo/tc-dev
_______________________________________________ tc-dev mailing list [email protected] http://lists.terracotta.org/mailman/listinfo/tc-dev
