Memcached 1.4.10 improves performance
In a release focused on improving thread scalability and performance, the developers of Memcached, the distributed memory object caching system, say that version 1.4.10 can now "feed data back faster than any network card can support". The performance enhancements saw developers report batched multiple key fetches per second rising from 1.6 million keys/second to "around 3.7 million keys/sec" on a quad core system with between 3 and 6 worker threads; more than six worker threads reduced speed, while a system with more cores was able to reach six million keys/second.
Tests which didn't involve the memcached bulk operators were limited by the throughput of local host or the network card. According to the developers, older versions of memcached are "plenty fast for just about all users" and this set of changes is about giving them more flexibility when they move to add features in the future. It also doesn't hurt that it improves memcached's latency on busy systems. The developers would like to hear from anyone who sees a regression from the changes in case they missed any cases in their testing.
Originally developed by Brad Fitzpatrick for LiveJournal, Memcached allows developers to aggregate the memory of server nodes in a system to act as a short term shared memory for the applications they run. It is used by sites such as Twitter, Flickr, Wikipedia, Youtube, Digg, Wordpress.com and others to help alleviate database load and make applications more responsive.