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

Re: which is faster ? ext3, reiserfs, xfs, jfs




On Mon, 5 Jul 2004, Kirk Strauser wrote:

> 
> > here's the fun freebie script to test which filesystem is faster
> > 	- format the disk(partitions) once
> > 	- do 3 passes copying 2.3GB of files from /dev/hda to /dev/hdc
> 
> Your benchmark is fundamentally skewed.  It uses tar to copy files from your 
> root dirctory to $MNT without caching the intermediate tarballs.  This means 
> that in each case, you're also measuring the read speed of your normal 
> filesystems in addition to the write speed of your target.  In the first 
> pass, you're also timing the speed at which your filesystems can fill their 
> directory listing caches, which will artificially improve subsequent runs.

thanx for looking at the "1st draft"

yes, on the cache issue .... i have not figured out how to clear the
cache between each pass 
	- umounting adn remounting it in the pass seems to help
	when testing thruput of 10MB.file.tgz across a wireless link

	- in the wireless tests, the 1st copy takes say 10minutes, but all
	subsequent tests copied "data across the wireless link" in
	5 seconds :-)  ( unmounting and remounting the seems to make
	each subsequent copy to come from the originating source instead
	of the local cache on the target )

another few things that will throw off the "tests"
	i have 1GB of memory ...  and the data files is a mere 2.3GB,
	which tells me the cache will have a big impact on thru put
	vs the "performance of the target"

	since i'm always copyig the "changing" master files, each
	subsequent copying will have a few more lines in /var/log of
	new data that wasnt there previous 
	( but compared to 2GB of data, a few lines shouldn't hurt any )

	another tests to do is to run a "set of benchmark test"
	on the target's filesystem instead of using the same
	"master disk" to do all the tests onto the target

i have the 3-pass numbers ... but since with 3 passes, that's not enough
info and cache flushing will be another day's fun

- anyway .. it was a fun project for a few hours

- will re-run it for a 10pass run at 10 minute each on 5 filesystem types
  ( 500 minutes ) will put me to sleep

c ya
alvin



Reply to: