[mythtv-users] SchedulesDirect timing out during Mythfilldatabase
mythtv at rtr.ca
Mon Apr 18 13:29:58 UTC 2011
On 11-04-17 10:10 PM, DeKay wrote:
> I know this. But I also mentioned that I suspect a bug in my kernel
> that is causing ext4 to do excessive journaling that is slowing down
> the database insertions. iotop is showing the ext4 journalling task
> being responsible for over 99% of the I/O at times.
Well, not a bug, but you could be onto something there.
Recent kernels (past year or two) really like to choose the slowest
possible disk I/O parameters -- the fs devs have been bitten by
the "data reliability even during an earthquake" affliction.
So this means the write-cache for your hard disk is probably "off",
which is a very stupid thing to do. Check for that with "hdparm -W".
Next, the ext4 filesystem may be mounted with "data=journal"
rather than "data=ordered". You can find out by looking at /proc/mounts
But in the end, it just might be mysql doing an excessive number of "fsync()"
or "fdatasync()" calls. Databases like to do things that way, but it does
mean they run very slowly when updating stuff.
I really should look into this mythfilldatabase slowness -- my system here
generally is pretty quick about it all, but it does have a blindingly fast
SSD for the root filesystem (and database).
But a clone of my system, installed for a buddy on mechanical media,
is much much slower during mythfilldatabase. I've never looked into it much,
but it could be the drives, the write-cache, or maybe his AMD CPUs are simply
that much slower than the 2.6GHz Core2 duo in mine ??
More information about the mythtv-users