Hi Sebastian, Does this mean that all controllers would write to a single backend? If so, this would not work-each Sequoia controller assumes it has attached backends to itself.
What you can theoretically do is run writes through Sequoia and read directly from the backends. This avoids a hop through middleware, which improves the cost/benefit for performance. Native databases are very fast, especially for things like key look-ups when data are largely resident in memory. However, if you do this you must ensure that you only look up data on enabled databases; otherwise, you risk queries on data that are out of date or even missing if the backend is being restored. Also, this works only on database implementations that do not take locks for read-only queries. Older databases based on the Sybase architecture (including MS SQL Server prior to the 2005 release) use read locks rather than MVCC to implement isolation of reads. These would interfere with Sequoia's ordering of writes to the database and cause potentially serious data corruption. Coming back to your architecture, it has some very likable properties for performance as well as some interesting availability properties-you can continue to serve reads out of cache, for example, even while the backend database is off-line or "stuck." If you are using Java you might want to check out EHCache as well as Sequoia. Meanwhile, thanks for your interest! Cheers, Robert On 11/1/08 5:08 AM, "Sebastien Pastor" <[EMAIL PROTECTED]> wrote: Hello the list, I am looking for a way to improve DB cache in a EC2 environment. The idea being : each ec2 instance has got a local "cache process" ( a bit like memcached) for reads and use the same process to handle writes; This would be resulting in a quicker response time for reads AND writes as the response time to reach the local process would be instant and it would be the one dealing with the latency between it and the DB backend. Do you think this could be achieve with Sequoia ? Basically the configuration i was figuring out would be : multiple controllers (one on each Ec2 instances) and one backend DB on a different instance. Thanks in advance ! -- Robert Hodges, CTO, Continuent, Inc. Email: [EMAIL PROTECTED] Mobile: +1-510-501-3728 Skype: hodgesrm
_______________________________________________ Sequoia mailing list [email protected] https://forge.continuent.org/mailman/listinfo/sequoia
