File System Timing Comparisons
 |
Kelsey Bjarnason
As a follow-up.
Fired up Ubuntu Live. Formatted a parbreastion as ext2, reiser (3.6) and xfs, and did some quick testing on each, using both Nautilus and Konqueror. Here's the results.
Nautilus took forever, over 200 seconds, regardless of file system, even for 10,000 files. It got the file count in a second or two, but wouldn't display a single entry.
Hello Gaming, buttembly and Linux Fans! :) Version 3.20 of HLA Adventure has just been released for both Linux and Windows: and HLA Adventure was written in Randall Hyde's HLA (High Level buttembly) programming...
Konq, by contrast, took a few seconds - about 5 - to both get the correct count and display the first page of items, though it continued to crank for 100+ seconds, even on 10,000 files - again, regardless of file system.
However...
Switching Konq to detailed list view, 10,000 files were displayed, with full information, in about 5 seconds. This suggests that Konq's slowdown isn't really filesystem-related, nor in handling largish numbers of items, but, rather, in generating the icons for them.
Either way, NTFS + Explorer is, in fact, a *hell* of a lot faster, if you're dealing with GUI tools and largish numbers of files. That said...
Creating the files was *blistering* fast in Reiser; about 1 second. So was reading, also about one second. As was deleting, again, about 1 second. It took 5 seconds in ext2 and 7 in xfs to create the files, though reading and deleting were again about a second.
Long and short of it seems to be that while the filesystem does, indeed, have an impact, what really bogs things down is how you choose to display the information. A simple ls is lightning fast. Working with the files from an application designed to process them - i.e. to do some actual work - is likewise fast. Trying to display them as "pretty pictures" can take agonizingly long times, regardless of the system, though Explorer is, in fact, a clear winner here.
What would be interesting to see - I didn't have the patience for it at the moment - is how Konq falls off with larger numbers. That is, if it follows an essentially linear progression as the number of files increases, or if, like Explorer, its time grows all out of proportion to the actual data set.