Wednesday, September 19, 2007

The ReiserFS / EXT3 comparison - more details?

After linking the thread in a chat channel I was asked by one of the admins for further details on the ReiserFS and EXT3 comparison mentioned in the previous blog flood. I have stated before that I am not a coder. I understand the concepts of coding, and I know how to debug written code, but I'm generally helpless at writing code from scratch and at even simple tasks like compiling. I freely admit this.

My involvement with the ReiserFS and EXT3 comparison came about more because of the hardware I immediately had on hand and was able to get access to. The real problem with benchmarking filing systems is that there really isn't anything tangible to work with. There are tons of programs available that will benchmark graphics and processors ranging from Professional OpenGL and Direct3D suites, to tools built right into commercial games, to home baked applications that record the frame rates.

There are a few tools available to benchmark Hard-drive performance under Windows, like WinBench from Ziff Davis, File Copy Test from Xbit Labs, and PC Mark. At the time of our testing we were only aware of one benchmark that existed for a Linux base, Iometer found at

This of course, in our view, wasn't exactly going to be a proper test then. So we had to come up with our own methods.

Our first tests then involved boot times, some of the results were listed here :

So um.. yeah. Might as well admit it, the What Can you Use post was actually a result of our tests. Video output gave us something that no other benchmark would. We could get an accurate snapshot of the time it took to load up an OS, copy files, delete files, run an FSCK, and so on without having an external program running on top of our host OS's that could affect the performance.

Our actual copying of the files used a couple of different metrics.

One of our tests involved creating 10 different folders and pasting the contents of the Chrono Symphonic MP3 files into each of the files. At about 108megs, this gave us over a gigabyte of files that were an average of 4 megs in size.

We did the same with the FLAC files. At about 455megs, this gave us well over 4gigs with an average file size of about 18megs.

This, we felt, was representative of the average limited computer user who rips CD's. We then proceeded to go through various procedures of burning the files to a disc, using Amarok to manage the files, as well as deleting and copying the files to external USB drives, as well as to a RAID 0+1 setup on an Adaptec PCI 32-bit SCSI card with 4 Drives (don't ask me who made them, I want to say WD although I probably remember incorrectly) in a RAID 0+1 stripe/mirror.

The thing was, with REISER4 used as our default format, or EXT3, we kept running into lots of other bottlenecks... the speed of the Optical drive, the speed of the PCI bus, the network interface speed, the amount of RAM in use, the processor...

But in our average desktop use we couldn't find any major performance difference either way.

Now, I'm sure that had we used more advanced server-level hardware, such as 64bit PCI Slot, or a PCI-X, or a PCI-Express attached SCSI card, or some other higher end setup, we may have seen a difference in our performance tests.

On bog-standard user-purchasable hardware that we could buy off of Newegg at that point and time and use in our computers? REISER4 wasn't any better than EXT3. It wasn't any worse... but, that's not how REISER4 is presented.

No comments: