Open Source Solutions - Latest comments on MuleCon2008 Users Day 2

John Davies [Visitor] in response to: MuleCon2008 Users Day 2

John Davies [Visitor] — Thu, 03 Apr 2008 17:10:04 +0000

In answer to the question above, it's quite complex but in essence because we get such high performance from in-memory "databases", we can start to serialise many of the transactions. Things that used to take 20 minutes now take a second or two and can usually be serialised. Transactions are not alleviated all together but we can vastly reduce the number of transactions and the impact of them on the data, this helps to further increase performance.

-John-

Joseph A. di Paolantonio [Member] in response to: MuleCon2008 Users Day 2

Joseph A. di Paolantonio [Member] — Wed, 02 Apr 2008 22:44:09 +0000

I just finished speaking to John about this. Here's the basics. In a clustered environment, write the transactions as an XML blob to memory, using something like GigaSpaces and index it with something like MAPS. When necessary, convenient, practical, or whatever other criteria makes sense for you, persist the XML Blob by writing to disk. The banks are using openSolaris, so the file system is ZFS, which branches changes much like subversion, thus keeping an history of any changes and ensuring that the history is maintained. By cutting out the database and the huge overhead of the ORM layer, the performance increase is amazing. This is very broad brush. You can get more details by John Davies and sorting through the 137,000 results. ;-)

Dimitar [Visitor] in response to: MuleCon2008 Users Day 2

Dimitar [Visitor] — Wed, 02 Apr 2008 21:58:02 +0000

Could you expand a bit more about how the "so great performance" alleviates the need for transactions?

Are we talking about cluster-durable vs. database-durable?