Infobright Blog

There seems to be a never-ending debate about what’s going to win out in the data warehousing ‘wars’ (I hate that term, but it’s the most apt for the topic at hand):  MPP or SMP.
I’ve seen lots of blogs, thoughts, and the like that seem to point both ways.  At Infobright, we’ve chosen to take the SMP approach in our offering.  Doesn’t mean we won’t put out an MPP solution at some point (and I’m confident that we could do this in a timely manner); just that at this point in time, we don’t see a huge burning need to move forward with an MPP architecture.

Not to say that there isn’t a place for MPP (just like we feel there is a place for SMP).  Business issues where I definitely think that there is a place for MPP:

•    Huge volumes of data (we’re talking well over 50 TB) where traditional partitioning methods won’t work.
•    Split second response times – for Operational Data Stores where fast query responses directly impacts revenue.  And those queries need to be predictable!

Of course, MPP comes with it’s own set of baggage:

•    Expense.  It’s not cheap to add in all of the redundancy and management necessary to keep an MPP architecture up and running.  Then you may have to add your own management and monitoring tools (at cost) on top of the vendor costs.
•    Configuration.  Although they’re getting better, the effort required to configure a solution on an MPP offering is sizeable and finding the resources to do it can be a challenge (great for those selling an MPP solution; not so great for those buying one).

Having spent considerable time ‘owning’ a data centre, I’m fairly prejudiced towards the SMP approach – the less cogs in a solution, the less likelihood of failure, and the less time and attention required.

Now, SMP has it’s own limitations:

•    Higher end of data storage, with Infobright’s compression, is around 30 to 50 TB (assuming that you’ve plugged a SAN into the configuration).  For a lot of other solutions, figure on 3 to 5 TB at most.
•    Less robust failover of any component – basically, if a box fails, the system fails – you need to switch everything over to a backup, which will result in some loss of performance.  Most MPP solutions have redundancy at a component level, resulting in a more robust failover.  And it’s built in.  With SMP you have to come up with the solution.

Of course, I like the SMP approach; it’s simple, easy to install, cost effective, and usually runs off of industry standard, readily available software.

Now, SMP and MPP are irrelevant if the solution doesn’t do what you want it to do. 

More on that later…