Re: [patchset] Lockfree fd lookup 0 of 5

[Dipankar Sarma <dipankar@in.ibm.com>]

On Tue, Aug 03, 2004 at 03:05:55PM +0530, Ravikiran G Thirumalai wrote:
> On Tue, Aug 03, 2004 at 02:53:17PM +0530, Ravikiran G Thirumalai wrote:
> > I ran tiobench on this patch and here is the comparison:
> > 
> > 
> > Kernel		Seqread		Randread	Seqwrite	Randwrite
> > --------------------------------------------------------------------------
> > 2.6.7		410.33		234.15		254.39		189.36
> > rwlocks-viro	401.84		232.69		254.09		194.62
> > refcount (kref)	455.72		281.75		272.87		230.10
> > 

Hm...

11514     6.7783  fget_light (vanilla)
13168     7.7224  fget_light (rwlock)
1993      1.2633  fget_light (kref)

Total ticks -

169886  (vanilla)
170520  (rwlock)
157760  (kref)

Of the 12126 ticks that were reduced by kref, 9521 came from 
reduction in fget_light(). So, lock-free fget_light() does help.
Also, it seems the lock contention is not much of an issue -

1203      0.7082  .text.lock.file_table

That explains why rwlock didn't help. I guess we are benefiting
mostly from avoiding the cacheline bouncing and removal of the
lock acquisition.

Thanks
Dipankar