I recently returned from Tech Field Day 15 where DataCore was giving the delegates an overview of their MaxParallel solution for SQL server. DataCore has a long established history in dealing with getting more performance from less with such products as SANSymphony and Hyperconverged Virtual SAN by leveraging Parallel I/O technology. At a very high level, this means that I/O requests in multi-core processor systems do not need to wait for the one core to complete a task before moving onto the next. MaxParallel for SQL server builds on this technology to deliver up-to 3X quicker response and 60% more transactions processed.
How does it work?
The exact mechanics of how this works is a closely guarded secret, I believe it is patent pending but don’t quote me on that. The solution though installs as an application on Windows server and runs as an upper-level filter driver. This allows MaxParallel to intercept the I/O requests and redistribute them in a parallel fashion, thus speeding up SQL transaction processing time. There is no waiting for one I/O queue to complete before moving onto the next queue. Word on the grapevine is it leverages a Least Recently Used caching mechanism to evict old data that does not need to be processed. Now I did pose the question, how does this solution identify SQL specific workloads, it doesn’t. In reality, this software would likely speed up processing of any application that can leverage multiple CPU cores. DataCore’s go to market strategy though is to target SQL server workloads as its a real-world problem for a lot of people and according to DataCore, represents a $20 Billion dollar market.
I spoke to DataCore again at IPExpo Europe, to see if I could get any more info about how the product works but alas I was given the same information. The standard test to show that the software is doing something though is to run a side by side comparison of the HammerDB SQL benchmarking tool. The demo on show was running on SuperMicro tin with a dual blade enclosure. Both were running the HammerDB test, one of them was optimised with MaxParallel for SQL server. On the face of it, the optimised server was performing a lot more transactions per minute so it does look like a useful tool.
Now I could paste a bunch of diagrams etc to depict the parallelism but there is not a lot to show, you can take a look at the datasheet for those diagrams and for a bit more info on the solution.
Why not check out the videos recorded from Tech Field Day 15 to get an idea of how the product is being positioned and how it works at a high level.
Videos Coming Soon
The cover your ass bit.
As this blog post is on the back of a Tech Field Day event, it would be fair to state that my travel, accommodation, food etc was covered for the duration of the event. The event is paid for by the vendors sponsoring the event. Opinions, however, are my own.