Facebook Announces WebScaleSQL Upgrade For MySQL

The WebScaleSQL branch of MySQL built by Facebook, Google, LinkedIn, and Twitter is about to get an upgrade. Facebook broke the news on Monday while announcing it’s taking a memcached tool open source and spearheading a new TODO group aimed at promoting open-source development.

Introduced to the world in March, WebScaleSQL is an open-source, Web-scale branch on top of Oracle’s publicly available MySQL Community Edition. The brand was developed by partners Facebook, Google, LinkedIn, and Twitter, all of which are big users of MySQL and all of which had been struggling with scaling the open-source database for their massive, Web-scale needs.

WebScaleSQL is distributed on Github as a set of patches on top of Oracle’s publicly available MySQL Community release. Features include a super read-only tool and the ability to specify sub-second client timeouts, buffer pool flushing improvements, optimizations to certain query types, and support for non-uniform memory architecture (NUMA) interleave policies.

[Want more on WebScaleSQL? Read Facebook Debuts Web-Scale Variant Of MySQL.]

The WebScaleSQL release to be available later this week will upgrade an existing Global Transaction ID feature in MySQL 5.6 that isn’t living up to Web-scale needs for replication and high availability at scale, according to Jay Parikh, VP of infrastructure engineering at Facebook.

“Global Transaction ID is a base feature, but we found, as we were trying to run it at Facebook, that there were several things that weren’t worked out in terms of being able to run it at the kind of scale that the [WebScaleSQL partners] run at,” said Parikh. “We’ve worked on bug fixes, making it faster, and making it more operationally sane to run at scale.”

Facebook also announced that it is taking a memcached tool it developed called mcrouter into open source. Mcrouter is a memcached protocol router that Facebook uses to handle “all traffic to, from, and between thousands of cache servers across dozens of clusters distributed in Facebook data centers around the world,” according to a statement from the company. The tool is able to handle nearly 5 billion requests per second at peak loads, according to Facebook.