Re: [patch 0/5] [PATCH,RFC] vfs: per-superblock unused dentries list (2nd version)

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

 



On Thu, Jun 01, Andrew Morton wrote:

> > Discussed in this thread:
> > 
> > http://marc.theaimsgroup.com/?l=linux-fsdevel&m=114890371801114&w=2
> > 
> > Short summary of the problem: due to SHRINK_BATCH resolution, a proportional
> > reclaim based on "count" across all superblocks will not shrink anything on
> > lists 2 orders of magnitude smaller than the longest list as tmp will evaluate
> > as zero.  Hence to prevent small unused lists from never being reclaimed and
> > pinning memory until >90% of the dentry cache has been reclaimed we need to
> > turn them over slowly. However, if we turn them over too quickly, the dentry
> > cache does no caching for small filesystems.
> > 
> > This is not a problem a single global unused list has...
> 
> Reasonable.  Whatever we do needs to be fully communicated in the comment
> text please.
> 

Yes, you are right. As I expected that this isn't the final patch I was a
little bit too lazy. Will do that for the next version.

> > > In particular, `jiffies' has near-to-zero correlation with the rate of
> > > creation and reclaim of these objects, so it looks highly inappropriate
> > > that it's in there.  If anything can be used to measure "time" in this code
> > > it is the number of scanned entries, not jiffies.

Ouch! Totally missed that. The measurement should be kind of round-based
instead.

> Don't do a divide?
> 
> 	sb->s_scan_count += count;
> 	...	
> 	tmp = sb->s_dentry_stat.nr_unused /
> 		(global_dentry_stat.nr_unused / sb->s_scan_count + 1);
> 	if (tmp) {
> 		sb->s_scan_count -= <can't be bothered doing the arith ;)>;
> 		prune_dcache_sb(sb, tmp);
> 	}
> 
> That could go weird on us if there are sudden swings in
> sb->s_dentry_stat.nr_unused or global_dentry_stat.nr_unused, but
> appropriate boundary checking should fix that?

  if (tmp) {
     sb->s_scan_count -= count;
     sb->s_scan_count -= sb->s_scan_count ? min(sb->s_scan_count, count/2) : 0;
     prune_dcache_sb(sb, tmp);
  }

  if (!sb->s_dentry_stat.nr_unused)
     sb->s_scan_count = 0;

In a normal situations, s_scan_count should be zero (add count and subtract it
again).
s_scan_count is increasing when we don't prune anything from that
superblock. If we finally reach the point where the s_scan_count is that high
that we actually prune some dentries, we slowly (count/2) decrease the
s_scan_count level again.
If the superblock doesn't have any unused dentries we reset the s_scan_count to
zero.

So s_scan_count is some kind of badness counter. I hope that this will still
be good enough for you, David.

Jan
-
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[Index of Archives]     [Linux Ext4 Filesystem]     [Union Filesystem]     [Filesystem Testing]     [Ceph Users]     [Ecryptfs]     [AutoFS]     [Kernel Newbies]     [Share Photos]     [Security]     [Netfilter]     [Bugtraq]     [Yosemite News]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux Cachefs]     [Reiser Filesystem]     [Linux RAID]     [Samba]     [Device Mapper]     [CEPH Development]
  Powered by Linux