August 2012 Summaries

MemSQL is an in-memory relational database that uses a unique architecture to achieve high performance and low latency. Its architecture combines the benefits of several technologies, including MVCC (Multi-Version Concurrency Control), lock-free coding, and code generation. MemSQL's use of an in-memory storage system allows it to eliminate the need for disk I/O, reducing latency and improving overall performance. The database also uses a proprietary algorithm to manage concurrency, which enables it to scale horizontally without sacrificing performance. Additionally, MemSQL's code generation capabilities allow developers to quickly build and deploy applications on top of the database, making it an attractive option for businesses looking to improve their development velocity.

SingleStore is a MySQL-compatible drop-in replacement designed to run fast in memory. A machine with 64 cores and 512 GB of RAM, similar to the Peak Hosting offering, was used to load half a billion records into SingleStore. The experiment aimed to compare loading efficiency between SingleStore and MySQL, using tricks such as multi-inserts, parallel loading, increased granularity, and a Python multiprocessing library-based load script. The results showed that SingleStore loaded data almost 200K inserts per second with a memory footprint of 267 GB, significantly faster than MySQL's 536 minutes to complete the same task.