MemSQL - Starting up / worth the move "Worlds Fastest Database"
MemSQL uses in-memory database technology. At the moment, db engines exist that can be configured in cluster, load balance and slave/master replication setups to achieve great speed and cost. Example, running 30x servers at 256mb and 1x server at 512mb as the master and have them all in replication and load balancer traffic spikes would more cost effective then 8gb setup costs for cloud providers - this is just heresay example. Clearly 256mb is an example.
Interesting the claim of worlds fatests database. IBM WebSphere Commerce customers use DB2 - IBM also have IBM solidDB - an in-memory solution:
IBM solidDB-Fastest Data Delivery
In-memory, relational database software for extreme speed
IBM solidDB® is known worldwide for delivering data with extreme speed. There are more than 3,000,000 deployments of solidDB in telecommunications networks, enterprise applications, and embedded software & systems. Market leaders such as Cisco, HP, Alcatel, and Nokia Siemens rely on it for their mission-critical applications.
IBM solidDB provides an ideal platform to build in-memory, real-time, enterprise applications that include universal caching. Compatibility enhancements make migrating applications to solidDB easy. Enhanced scalability and transparent failover with solidDB lowers overall total cost of ownership. Improvements in the latest release bring more reliability and faster performance to key OLTP workloads with the solidDB Universal Cache feature.
IBM WCS and IBM is enterprise and expensive, thought be interesting showcasing a major company already doing in-memory so surely not new thing.. where are the memsql customers and staff and also where are the DB experts on SO - please don't mark your own answer as the answer too..
The MemSQL lacks community support, we have no information about the enterprise license (there is available a version for more than 32GB) and maybe there is no proof of speed for your application yet.
However you have to try it. Really. The performance for static queries and prepared statements are promising, the best use case I could emphasize is for low latency writing/updating records at extreme concurrency. You can't achieve performance like this with relational databases and the NoSQL solutions with this performance are rare.
In short time it will get SQL92 compliency, so the development and testing will be easier for you or for your developers (opposed to a NoSQL DB backend). There are thousands of applications with built-in performance benchmarks and long-running stability testing, choose the one which matches your case most. Personally I've tested Drupal, the worst performance was the same performance as a heavily customized MySQL configuration, on average it became 10x faster on DB side for logged in users.
About the performance issue - Ask yourself the following question:
If you took Percona MySQL (high performance MySQL server) and you put it on a multi-core machine with 512GB RAM, storing the databases in a ramdisk, and let's assume this is more than enough space to hold your databases entirely in memory, Would it still be slower than MemSQL on the same hardware? The only way to test this is to compare both of them.
About the pricing tiers, A company does not provide prices on its website when it understands that it can price it differently for different companies. For example, they will find a way to get General Electric to pay $500,000 for it, but a smaller startup will be able to get away with paying "only" $10,000 for the same product.