[mythtv-users] Source of my IOBOUND problem?

Blammo blammo.doh at gmail.com
Tue Jun 12 18:25:09 UTC 2007


On 6/8/07, Craig Huff <huffcs at yahoo.com> wrote:
> Boy,  having been a Solaris sysadmin for many years just doesn't help when
> dealing with Linux.  -- so many things I never had to deal with --

Trust me, even having been a linux admin for a bunch of years didn't
prepare me for the kind of IO load a serious multimedia cluster can
require. It's called break it till it works like you want.. :)

> IO Scheduler:  How do I determine what I've got running, how do I find what
>    alternatives I already have available, how do I switch to one of those, and how
>    do I go about getting the "deadline" scheduler if I don't already have it as part
>    of my FC6 installation?

Default, as I recall, is anticipatory.  Here's a link that explains
them decently:

http://www.wlug.org.nz/LinuxIoScheduler

I was corrected on this list once before for my incorrect explanation
of deadline, so let me give you my unscientific one....

All other schedulers caused frontend-playback-stutter during high
disk-IO except 'deadline', for me. Your mileage may vary.. :)

You can change the scheduler used, at boot time, by adding a line in
your /etc/grub.conf.. here's what mine looks like:

title CentOS (2.6.20-myth-smp)
        root (hd0,0)
        kernel /vmlinuz-2.6.20-myth-smp ro root=/dev/hda2 elevator=deadline
        initrd /initrd-2.6.20-myth-smp.img

I also do a little bit of tuning on /etc/rc.d/rc.local.. setting
deadline again was redundant, but it helps incase I forget to add the
line after I upgrade a kernel.

blockdev --setra 1024 /dev/sda
echo deadline > /sys/block/sda/queue/scheduler
echo 1024 > /sys/block/sda/queue/nr_requests

You can tell which scheduler is running on a specific Device by
checking the following:

cat /sys/block/XXX/queue/scheduler

here's what mine looks like for my SDA (raid array)

[root at backend1 ~]# cat /sys/block/sda/queue/scheduler
noop anticipatory [deadline] cfq

> Thrashing and Disk Performance:  Doable, and have this plan in mind:
>    Install second disk to put all but /video partition on,
>    mount equivalent partitions under /mnt and run
>            # dump /current/hda# | restore /new/hdb#
>    until all migrating partitions are copied, edit /etc/fstab (keeping a copy of the
>    original!), and reboot (with my trusty Knoppix LiveDVD at the ready ;-)
>
>    If this fails, then it's on to software RAID striping SATA disks -- this is a good
>    excuse for expanding storage capacity AND dipping my toes in the SATA pool
>    for the first time, eh?

Just my 2cents, but for the money, it's realllly hard to beat the
benefits of hardware (not fake) raid. It's the destination that most
of us end up at after thrashing around with other options.. 3ware,
Areca, Adaptec, all make good HW raid cards. SW can work ok, but the
IO load can drop the overall stability of your box..I fought with
XFS/NFS/SWRaid5 for a long time before gave up on SWraid. Was the best
money I spent....

>    Also, I get the feeling I should be adding more RAM.  Currently I have 512MB.
>    Recommendations, anyone?

More RAM, at least for the backend, means more disk-cache, which helps..

> Commercial Flagging:  Ummm...How?
>    I know how to schedule jobs with cron and anacron, but how do I restrict what
>    times mythcommflag CAN run?

mythtv-setup

there's a page (I think it's jobqueues) that lets you set low-med-high
priority on comm-flag jobs, and what times of day/night you let them
run. The setup is not global, but rather specific to each machine you
run mythtv-setup on.

hope this helps.


More information about the mythtv-users mailing list