Niclas Hedhman wrote: > Does this at all help on the "redundancy" in storage? > > If Entity1 lives on NodeA, then it can't only store itself on NodeA > (assuming NodeA is a single host) as it will get destroyed both in RAM > and disk at the same time. Destroyed on disk? What is the scenario here, permanent machine failure? Assuming that you're addressing failover, right? > But would it be possible to create small > clusters which shares a redundant store for storage, but where each > Entity only live on a single host? Comparing to GigaSpaces infrastructure, there is one service that resembles this. They call it mirror service which a cluster wider service that stores stuff asynchronously. Asynchronously because otherwise it would be a bottleneck just like a traditional database would be. The mirror service typically stores writes in the spaces, so it's not exactly what I we discuss here. (You can of course confine mirroring to relevant entries)
So, the centralization comes at the price of a small inconsistency with data in memory. The nodes being mirrored employ a simple acknowledgement protocol to keep track of what data has been mirrored, combined with a retry scheme. > Is that at all smart? And isn't the > "share nothing" philosophy creating new (interesting) challenges for > indexing and searches? (Or does one just accept that Google, Yahoo... > are "almost-infinite scalable") > /Niklas > Cheers > Niclas > > _______________________________________________ > qi4j-dev mailing list > [email protected] > http://lists.ops4j.org/mailman/listinfo/qi4j-dev > > _______________________________________________ qi4j-dev mailing list [email protected] http://lists.ops4j.org/mailman/listinfo/qi4j-dev
