[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: how to counteract slowdown

"Stephen C. Tweedie" wrote:
> Hi,
> On Mon, Nov 19, 2001 at 01:01:32PM -0800, Andrew Morton wrote:
> > Until it's empty, I'm afraid.   If we determine that there is
> > not sufficient room in the journal we write out _all_ checkpointable
> > data.
> At least that's something we can fix.  My old 2.2 code would walk
> through transactions on the checkpoint list in order until there was
> enough free space (the "nblocks" argument to log_do_checkpoint).
> Andrew's checkpoint streamlining seems to have lost that, but we can
> restore it if we want to.

umm..  `nblocks' is unused in 0.0.7a as well.

> Whether we _do_ want to is a different matter, as we'd risk trickling
> out so slowly that the disk would remain permanently busy, albeit with
> a short enough write queue that latency wouldn't be too bad.  I'll
> have to experiment with this...

yes, writing less stuff simply _has_ to lower throughput.

It may be better to start IO on all of it, but to only
wait on as much as we need.


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]