Re: Fragmentation of Journaling FS

[Utz Lehmann <leh@de.tecosim.com>]

Hi Constantin

Constantin Loizides [Constantin.Loizides@isg.de] wrote:
> > I made a quick and dirty test running this:
> > 
> > while time cp -a /usr/src/linux/drivers/ /mnt/xxx-`date '+%s'`; do sync; \
> > df | grep mnt; done
> > 
> Ok, will try this when my test system is available...
> 
> Please notice, that I am using a different setup:
> different version of xfs,
> different file size distribution (i checked that on my /drivers/ dir)
> 
> Can you check your test on a small partition of 512 MB, just to 
> make sure, that it gives the same flatness?

Ok, i have run my test on a 1GB and 512MB volume (using lvreduce).

1GB volume:

user	system	elapsed	CPU	Used	Avail.	Use%

0.10	2.55	0:22.01	12%	75100	968676	8%
0.17	2.98	0:22.92	13%	150040	893736	15%
0.09	2.62	0:22.63	11%	224996	818780	22%
0.14	2.77	0:22.51	12%	299952	743824	29%
0.08	3.01	0:23.24	13%	374892	668884	36%
0.10	2.92	0:22.77	13%	449832	593944	44%
0.18	2.89	0:22.95	13%	524772	519004	51%
0.07	3.10	0:22.88	13%	599668	444108	58%
0.09	3.11	0:22.95	13%	674592	369184	65%
0.12	2.78	0:23.20	12%	749516	294260	72%
0.14	3.02	0:23.92	13%	824456	219320	79%
0.14	3.11	0:24.03	13%	899412	144364	87%
0.14	3.29	0:23.91	14%	974304	69472	94%


512MB volume:

user	system	elapsed	CPU	Used	Avail.	Use%

0.12	2.60	0:16.69	16%	75100	444388	15%
0.16	2.68	0:17.29	16%	150040	369448	29%
0.09	2.72	0:16.96	16%	224996	294492	44%
0.11	2.61	0:16.62	16%	299904	219584	58%
0.11	2.60	0:16.29	16%	374828	144660	73%
0.12	2.79	0:15.70	18%	449768	69720	87%


No sharp drop.
That the elapsed time is reduced by the volume size is very interessting. On
the 512MB volume it's only half of the 4GB volume. My test is very quick and
very dirty.


btw: For your agesystem you can maybe add following:
- random file deletion
- deletion of whole trees
- simultaneous writing of files (very important, a lot of filesystems are
  fragmenting these files badly)
- appending to files in different sized pieces (like logfiles)


utz