I’m impressed by a session called ‘memcached in mixi’ by Masahiro Nagano. He detailedly explained their effort to scale the Japan’s biggest SNS with memcached.
They are using more than 100 dedicated memcached servers for caching. All of machines has 4GB of memory with Pentium 4 or D, which are too weak for database servers now. They use Nagios (with check_tcp command) to check the servers alive or not.
Their most burdened machine processes 15000 request per second (400Mbps), and it’s still properly working with lower load-factor number.
Their load-balancing method is CRC(key) mod number_of_servers, so it’s almost impossible to add a server to memcached cluster because it causes a drastic decline of cache hit percentage. They plan to introduce new ‘consistent hashing’ method.
Mixi is using TokyoTyrant, a fast DBM engine created by Mikio Hirabayashi, to save login data such as last login time.
Michael Schwern gave a closing key note called ‘Perl Is unDead’. He argued against the perception that people think ‘Perl is dead’. His talk was very vibrant, and woke up my sleepy mind filled up with too much tech talks.
He proposed a solution that Perl users should take their own domains for their projects instead of uploading it to CPAN. Perl user is too much depending to CPAN, the big repository of Perl codes. Activities inside CPAN is virtually invisible from outside of Perl community.
YAPC::Asia 2008 was successfully held with 524 tickets sold. Almost half of the talks was in English. It’s a rare occassion to meet with world’s top notch engineers in Asia.
I’m not using Perl for 5 years, but I enjoyed YAPC::Asia so much. You shouldn’t miss the event next year!
See Also: (in JP)
* mixi->{engineers_blog}
Shunichi Arai