Bug 252384 - cmirror resync slows down when I/O is outstanding to the mirror
Summary: cmirror resync slows down when I/O is outstanding to the mirror
Keywords:
Status: CLOSED DEFERRED
Alias: None
Product: Red Hat Cluster Suite
Classification: Retired
Component: cmirror
Version: 4
Hardware: All
OS: Linux
medium
medium
Target Milestone: ---
Assignee: Jonathan Earl Brassow
QA Contact: Cluster QE
URL:
Whiteboard:
Depends On:
Blocks:
TreeView+ depends on / blocked
 
Reported: 2007-08-15 18:14 UTC by Nate Straz
Modified: 2013-09-23 15:31 UTC (History)
7 users (show)

Fixed In Version:
Clone Of:
Environment:
Last Closed: 2013-09-23 15:31:26 UTC
Embargoed:


Attachments (Terms of Use)

Description Nate Straz 2007-08-15 18:14:47 UTC
Cloning off the piece about performance of mirror resync dropping.

+++ This bug was initially created as a clone of Bug #252007 +++

Description of problem:

-- Additional comment from nstraz on 2007-08-13 15:40 EST --
We've run into similar problems with a fresh syncing mirror.  When we do a mkfs
on a not-yet-synced mirror, the mirror sync performance drops.  I did a quick
test now and found that on a 100G mirror with 3 legs (lvcreate -m 2), sync
performance was about 0.9% completion per minute.  After starting a mkfs -t gfs
(which hangs) mirror completion drops to 0.05% per minute.

-- Additional comment from jbrassow on 2007-08-13 15:46 EST --
Cluster mirrors in 4.5 did not have a way to coordinate device resyncing and
nominal I/O.  Therefore, nominal I/O had to be deferred until the regions were
resync'ed - causing excessive delay.  This has been addressed in 4.6.  See bug
238629.

Comment 1 Jonathan Earl Brassow 2007-09-28 15:28:00 UTC
The latest build in 4.6 (>= 9/28/07).  Addresses some of these concerns.  I was
forced to trade consistency for speed, however.  So, things will be faster than
4.5, but I'd like to know what the limitations are.

assigned -> needinfo


Comment 2 Jonathan Earl Brassow 2008-04-02 19:03:13 UTC
needinfo -> assigned

I have a patch ready, but it still needs to be tested.  I borrowed some ideas
from the rhel5 code, which I am using to limit the network traffic while there
is recovery + nominal I/O happening.

Perhaps this patch could make 4.8.

Comment 5 Lon Hohberger 2013-09-23 15:31:26 UTC
The Red Hat Cluster Suite product is past end-of-life; closing.


Note You need to log in before you can comment on or make changes to this bug.