Re: [PATCH 0/5] multiple block allocation to current ext3

[Mingming Cao <cmm@us.ibm.com>]

On Tue, 2006-01-10 at 21:25 -0800, Andrew Morton wrote:
> Mingming Cao <cmm@us.ibm.com> wrote:
> >
> > Tests done so far includes fsx,tiobench and dbench. The following
> >  numbers collected from Direct IO tests (1G file creation/read)  shows
> >  the system time have been greatly reduced (more than 50% on my 8 cpu
> >  system) with the patches.
> > 
> >  1G file DIO write:
> >  	2.6.15		2.6.15+patches
> >  real    0m31.275s	0m31.161s
> >  user    0m0.000s	0m0.000s
> >  sys     0m3.384s	0m0.564s 
> > 
> > 
> >  1G file DIO read:
> >  	2.6.15		2.6.15+patches
> >  real    0m30.733s	0m30.624s
> >  user    0m0.000s	0m0.004s
> >  sys     0m0.748s	0m0.380s
> > 
> >  Some previous test we did on buffered IO with using multiple blocks
> >  allocation and delayed allocation shows noticeable improvement on
> >  throughput and system time.
> 
> I'd be interested in seeing benchmark results for the common
> allocate-one-block case - just normal old buffered IO without any
> additional multiblock patches.   Would they show any regression?
> 
Hi Andrew, 
  One thing I want to clarify is that: for the buffered IO, even with
mutlipleblock patches, currently ext3 is still allocate one block at a
time.(we will need delayed allocation to make use of the multiple block
allocation)

I did the same test on buffered IO, w/o the patches. Very little
regression(less than 1% could be consider as noise) comparing 2.6.15
kernel w/o patches:

buffered IO write: (no sync)
# time ./filetst  -b 1048576 -w -f /mnt/a
	2.6.15		2.6.15+patches
real    0m25.773s	0m26.102s
user    0m0.004s	0m0.000s
sys     0m15.065s	0m16.053s

buffered IO read (after umount/remount)
# time ./filetst  -b 1048576 -r -f /mnt/a
	2.6.15		2.6.15+patches
real    0m29.257s	0m29.257s
user    0m0.000s	0m0.000s
sys     0m6.996s	0m6.980s


But I do notice regression between vanilla 2.6.14 kernel and vanilla
2.6.15 kernel on buffered IO(about 18%). 

# time ./filetst  -b 1048576 -w -f /mnt/a
	2.6.14		2.6.15
real    0m21.710s	0m25.773s
user    0m0.012s	0m0.004s
sys     0m14.569s	0m15.065s

I also found tiobench(sequential write test) and dbench has similar
regression between 2.6.14 and 2.6.15. Actually I found 2.6.15 rc2
already has the regression.  Is this a known issue? Anyway I will
continue looking at the issue...

Thanks,
Mingming