>>>>> "Duncan" == Duncan Hutty <[email protected]> writes:
Duncan> -----BEGIN PGP SIGNED MESSAGE-----
Duncan> Hash: SHA1
Duncan> On 9/11/12 9:01 AM, Edward Ned Harvey (lopser) wrote:
>> I'm being exposed to drbd for the first time, and I'm not
>> impressed. I'm finding it's a sort-of inflexible kludge for the
>> purpose of mirroring storage locally and remotely.
>>
>> What about md or lvm mirroring to an iscsi device? Experience?
>> Opinions?
>>
>> Ideally, I'd love to see something comparable to zfs mirroring with
>> an iscsi device. That is - If the iscsi device disappears and
>> reappears, it only needs to resilver the blocks that changed in the
>> meantime. And the iscsi device is never inconsistent. (Both
>> characteristics lacking by drbd).
>>
>> But I don't have a lot of experience with md or lvm mirroring.
Duncan> Just a single data point:
Duncan> Recently I had the opportunity to resilver a raid1 mirror that uses
Duncan> md. In its defence, I will say that there had been no attempt to
Duncan> optimise, configure or in any way improve the experience (is there
Duncan> anything that can be done to improve this?). It had just been created
Duncan> with:
Duncan> mdadm --create /dev/md0 --level=1 --raid-devices=2 /dev/sdb1 /dev/sdc1
Duncan> However, although resilvering the mirror while the device is active is
Duncan> possible, the performance was inadequate to the point of uselessness.
Duncan> After several hours it was killed at <40% complete. (Hardware was some
Duncan> modern quad core server with 96G and 15K rpm disks, otherwise more or
Duncan> less completely inactive at the time).
Did you ever look at the settings of
/proc/sys/dev/raid/speed_limit_max and cat
/proc/sys/dev/raid/speed_limit_min? Echoing new values in there will
give you a big speed up. By default, they're left low in older linux
kernel versions so as to not impact production.
Duncan> It was easier to copy the data (few 100s of GB) off the (1TB?)
Duncan> filesystem and rebuild the mirror from scratch. Perhaps my
Duncan> surprise was only due to my ignorance of the underlying
Duncan> mechanisms.
The other thing to do is to make sure there's a bitmap associated with
the array. You can do this with:
mdadm --grow --bitmap=internal /dev/md0
Good luck,
John
_______________________________________________
Tech mailing list
[email protected]
https://lists.lopsa.org/cgi-bin/mailman/listinfo/tech
This list provided by the League of Professional System Administrators
http://lopsa.org/
