On Tue, May 13, 2003 at 10:50:45AM -0500, Steve Lord wrote:
> > > By adding the periodic actvity thread we activate some code which
> > > looks for an empty AIL and writes out a dummy log record to record
> > > the new tail of the log.
> >
> > Does it check that it already wrote the marker before writing again ?
> > If not the laptop users will hate you ;)
>
> It does it twice, then stops
I would still suggest making the flush interval a sysctl and
provide the "SIGSTOP to stop thread until SIGCONT" interface. We did this
for ext3/jbd and it was very useful for laptops.
> > And I imagine changing this could add lots of seeking ...
> > (between the inode areas and the data areas)
>
> It may be more along the lines of holding off on letting the inode
> out to disk, but it would be complex.
I was thinking the same, but even with that it would add more seeking
I guess. Currently it would write out a lot of inodes in one go and then
later or earlier the extents. With extent dependencies that would be split
into multiple extent / inode writes with seeks inbetween.
-Andi
|