[DRBD-user] DRBD write throughput very very bad

Lars Ellenberg lars.ellenberg at linbit.com
Fri Apr 24 13:11:24 CEST 2009

Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.


On Fri, Apr 24, 2009 at 12:18:47PM +0200, Dominik Klein wrote:
> > ah. and btw. do not use "none",
> > that is to say, do NOT use "no-disk-drain".  in general, that setting
> > will violate write ordering constraints, and gives very little benefit.
> > it is useful only in very special circumstances.
> 
> Namely on known-to-work-well battery-backed-write-cache controllers. Is
> that about right?

nope.  on known-to-not-reorder-requests-and-no-volatile-caches-involved
backends. which are exceptionally rare.

in linux, you need to be on top of "elevator=noop",
and no driver involved, or SAN iSCSI/FC target (or sata on-disk
circuitry, for that matter) is allowed to even think about re-ordering.

if (and _only_ if) nothing in the stack re-orders writes, (and won't
lose writes on power loss or crash), then barriers become pretty much
no-ops (with possibly performance hurting side-effects), and any waiting
on already issued requests can be left out.

-- 
: Lars Ellenberg
: LINBIT | Your Way to High Availability
: DRBD/HA support and consulting http://www.linbit.com

DRBD® and LINBIT® are registered trademarks of LINBIT, Austria.
__
please don't Cc me, but send to list   --   I'm subscribed



More information about the drbd-user mailing list