In preparation for Open World, I asked some of our engineers to recreate a demo that Oracle has been using over the last year to show off their “in-memory option.” It’s impressive to look at: the demo shows the database searching through billions of records from Wikipedia search trend data for popular terms in less than a second.
The thing about Oracle’s demo is it runs on a gigantically expensive server. In fact it is the biggest one they have at 32TB of RAM and hundreds of CPU cores. They are proud to say that it costs “only” three million dollars. Here is a picture of the beast, which is bigger than your refrigerator.
Distributed databases that scale horizontally on commodity hardware are the way of the future. To prove that, we created a modest cluster on Amazon and built the same demo using MemSQL. The cluster consists of just 6 Amazon nodes.
We’ve been presenting our new demo at Oracle OpenWorld this week, and the responses have been incredible. People can search the Wikipedia data set for popular keywords and trends over time. The ideas they have come up with to search are amazing! And all of it runs on a 6-node cluster that costs about $5/hr on Amazon or less than $20k if you wanted to buy 6 servers outright. All in all it cost less than the t-shirts we gave away at the show.
One of the reasons that MemSQL’s version goes much faster for these queries is that it supports sorted columnar tables. Oracle In-Memory Option is a columnar compressed dataset with no ordering information. MemSQL’s compression format allows it to perform operations on compressed data, and take advantage of the sort order. Ultimately this demonstrates how elastic, cloud friendly databases provide a much lower cost of ownership compared to monolithic (literally) appliances.
Try MemSQL for 90 days: http://www.memsql.com/download/
MemSQL is an in memory database that enables businesses to process transactions and run analytics in real-time with SQL.