Are you using a separate read service or just the mogstored for reads
and writes?
mogstored's ability to serve files without timing some out occasionally
is a little suspect at the moment. A common practice is to fire up
lighttpd or apache on a separate port and set the 'http_get_port'
(you'll have to look up the actual mogadm command?) to point to that
other web instance. Then the mogstored only handles writes and
statistics for mogilefs, which it should be fine at doing.
-Dormando
Brian Lynch wrote:
Hi,
Does anyone have any experience tuning a larger installation of
MogileFS? We have approximately 70 million files distributed among 5
large storage nodes (46 devices each; 2 X Dual Core AMD; 16G memory). We
started experiencing a number of connection timeouts from the systems
hosting the tracker (mogilefsd) to the backend storage nodes
(mogstored). We’ve made the following tweaks to reduce the failures,
but haven’t been able to determine why mogstored is timing out. If
anyone has any ideas on tuning mogstored/perlbal or tweaking the overall
installation, that would be extremely helpful.
Thanks,
Brian
TIMEOUT CHANGES
/usr/lib/perl5/site_perl/5.8.8/MogileFS/Worker/Replicate.pm
71: Changed watchdog_timeout to 60
796: Changed Timeout to 50
822: Changed timeout to 50
/usr/lib/perl5/site_perl/5.8.8/MogileFS/Worker/Monitor.pm
88: Changed connection timeout to 10
