Today everybody writes about MySQL Conference & Expo and I am not an exclusion. I am under impression of count of storage engines were presented. In good old time when Oracle bought InnoDB, MySQL did one step – announced MySQL supports Plugginable Storage Architecture. In that time nobody was able to predict what is the outcoming of such feature, but currently we all see it.
Let me list Transactional Storage Engines I heard of on Conference and main impression about.
InnoDB On the conference InnoDB presented two new interesting features – transparent ZIP of pages stored on disk. Potentially it can imporove performance of tasks with intensive I/O load a lot. Second feature – fast index build. Both should be available in 5.2 MySQL.
PBXT Interesting engine, mainly because of way to store new records “never updates” and just write to the end of file. Surely it has as benefits as and drawbacks. We are going to test PBXT in write workload, though currently comparison with InnoDB will not be fair, as PBXT do not perform fsync() after commit and fully rely to OS. The developers promise to fix it soon.
Falcon The Storage Engine which is developed in MySQL. The performance we see in our previous benchmarks was not good enough, but there was some fixes in 5.2 tree. Also this engine will not support index coverage optimization, that makes Falcon less attractive from performance reasons.
Solid The engine with both pessimistic and optimistic locking on the table level. The interesting note Solid proposes High Availability solution for Solid Engine, which can be replacement of MySQL replication (the solution was only announced and will be available later)
And three engines which were only presented and details only available from marketing materials:
NitroEDB The engine promises to handle in very fast way aggregative functions on massive data sets. As I understood they use special index to store pre-calculated values for limited number of functions, and then engine can retrieve values from index. The interesting question if engine is able to calculate combination of functions or user-defined function on TB data. I would like to test it, but most likely it is not possible in near future, as license price is 20.000$ per server ( the price I heard somewhere, not sure if it’s 100% correct)
Infobright Infobright is non-transactional and read-only engine designed for DataWarehouse applications with compression rate from 10:1 to 30:1
ScaleDB The engine based on special index, so named “Trie”, the extension of “Patricia Tree” data structure. The developers promises ultimate speed and scalability for index access operations. It’s interesting to test it in real situations, because theoretical ideas can be very far from life. Also nobody prohibits to add such index to other storage engine if it will show good performance results.
The wide choice is always good, but I’m afraid making right decision about Storage Engine will be much harder.