On Fri, 2004-04-30 at 22:25, Dave Kleikamp wrote:
> On Fri, 2004-04-30 at 15:03, Bart Samwel wrote:
> > On Fri, 2004-04-30 at 21:52, Dave Kleikamp wrote:
> > > Bart, did you consider jfs along with the other file systems? 
> > 
> > Nope. I've never heard anybody asking for JFS support before, so I
> > haven't considered it.
> > 
> > About what laptop mode needs: if JFS doesn't initiate any write activity
> > at all unless blocks are flushed by pdflush, then you're safe. I'm not
> > quite sure what you mean by "current activity" though, so there might be
> > a problem depending on what that means.
> 
> All I meant by that is that if a thread were actively writing to jfs (or
> creating, renaming, deleting files, etc.) that that activity may cause
> older data & metadata to be written.  This is such an obvious statement,
> I really didn't need to say it.

Yeah, but I still don't quite get it. :) Are you referring to the
situation where there are too many pending changes and some older ones
need to be flushed, because of a journal size limit or something like
that? Or are we talking about when something is actually physically
forcing some changes to disk (using something like fsync()), which might
force older changes to disk because of ordering restrictions?

> >  Other journalling filesystems
> > all have some kind of "transaction lifetime limit", usually 1 to 5
> > seconds, which will commit a transaction if it lives longer than that. 
> > And committing means *writing*, at least to the journal. Can you give me
> > some more info?
> 
> jfs doesn't really have such a transaction lifetime limit.  It pretty
> much depends on pdflush to initiate writeout of dirty inodes, at which
> time outstanding transactions are committed to disk.

Ah, that sounds exactly like what Laptop Mode needs.

> > > Do you
> > > have any suggestions for making jfs behave better on a laptop?
> > > 
> > > I admit that I haven't played with laptop mode yet.
> > 
> > I think that should be the first step, before we go investigating if we
> > need anything more. I'll try and check it out, I have some standard test
> > scenarios so I'll follow up on this as soon as I've run them.
> 
> Thanks, let me know if I can do anything to help.

I executed my standard test scenario:

- Create JFS filesystem and mount it on /jfs
- Put my system to silence (turn of all daemons that regularly do disk
access)
- Enable block dump
- "echo > /jfs/testfile" (this gives me a "dirtied file" message from
block dump, and makes sure the root directory of JFS is cached)
- Wait for the system to quiet down and then do "echo > /jfs/testfile"
again
- Time how long block dump remains quiet.

JFS seems to be behaving very well with Laptop Mode: I got over two
minutes of silence until I gave up. Any of those other filesystems
(ext3, ReiserFS, XFS) would have already given up after a couple of
seconds. So, until I hear otherwise I'll assume laptop mode and JFS work
just fine.

--Bart
_______________________________________________
Jfs-discussion mailing list
[EMAIL PROTECTED]
http://www-124.ibm.com/developerworks/oss/mailman/listinfo/jfs-discussion

Reply via email to