Latest tests by Donn Edwards

[pYjamas]

What do you guys think about Donn Edwards latest tests of defragmentation software? Especially JkDefrag and Puran Defrag.

JkDefrag performs the same as the big names such as Perfectdisk, Diskeeper and O&O Defrag which is very good for being a free software but what surprised me the most was that Puran Defrag performed almost 10 % better on average than all the others.

Are you surprised that JkDefrag didn't perform better or do you think that the tests are incomplete and more testing is needed?
What are your thoughts on Puran Defrags result?

http://donnedwards.openaccess.co.za/2007/11/benchmarks-jkdefrag-328.html
http://donnedwards.openaccess.co.za/2007/12/benchmarks-puran-defrag-40.html

/pYjamas

[Lexar]

To let JkDefrag show its full potentials, we should specify in commandline arguments the options appropriate to how the drive is used. I guess Donn Edwards doesn't seem to have done a full test of JkDefrag.

[Myriades]

Hi all
Actual JKDefrag version is 3.33, so...
Regards, Laurent

[jeroen]

Are you surprised that JkDefrag didn't perform better or do you think that the tests are incomplete and more testing is needed?

I see that Donn has finally found a commercial defragmenter that performs better than JkDefrag. Commercial defragmenters should perform better than JkDefrag, as I have always said, because they have a lot more resources to spend on programming than a single old nerd like me. Also, JkDefrag does not yet support boot optimization. So the answer is no, I am not surprised.

[pYjamas]

Currently I'm using puran for the windows partition and jkd for the other partitions which are storage. When boot optimization and prefetching is incorporated into jkd hopefully it will perform the same or better than puran and i can throw puran in the trash

Keep up the good work jeroen. You'll get 'em soon enough!

[Lundholm]

Just some theoretical thoughts on these benchmark tests. I have not gone into much detail on this, but it seems that the test consists of measuring the read time for a large set of files before and after defrag, correct me if I'm wrong.

I don't use my computer to read a large number of files. I use my computer to read some files many times and some other files not so often. This is true for most users, I believe. We all read some Windows files and applications every day. Many Windows files and applications are rarely used, if ever. Mass file access times or average file access times are not very relevant for me (sorry Donn).

So I need something that can speed up the access to my favorite files, and I believe that JKdefrag sort option 9 does exactly that. From a theoretical point of view, the only improvement I can see, is the grouping of files used for a specific purpose, like boot/startup, and I believe that Jeroen is working on that.

Option 3 is absolutely brilliant, because it is so fast, and option 9 is even more brilliant, because it makes my system fast. No need to consider any other defragger.

Cheers

[Lundholm]

From a theoretical point of view, the only improvement I can see, is the grouping of files used for a specific purpose, like boot/startup, and I believe that Jeroen is working on that.

Sorry, but I forgot to mention the new sorting suggested by Esprit in this thread:

http://www.kessels.com/forum/index.php?topic=924.0

This is also a theoretical improvement.

Cheers

[Donn Edwards]

As a word of explanation about the tests: I used the version of JkDefrag available at the time (Nov 2007) of doing the tests, and used it without any additional flags, just the default settings.

The idea behind the tests was to measure the read time of most of the files listed in the layout.ini file, since these are files that are either accessed during boot time or are apps regularly used by the user. It was the closest I could get to finding an objective measure of the "speed" of the PC as experienced by the user.

The only reliable lesson I have learnt from these tests is that different programs have a different impact, and that you can't predict the result. I expected "bad" defrag programs like Diskeeper and Norton SpeedDisk to perform worse than "better" programs like JKD and PerfectDisk. This didn't happen.

It turns out that measuring the effects of a defrag program is not as simple as it looks, and I'll have to go back to the drawing board to figure out another method of measuring defrag programs.

For the record, I bought a copy of PuranDefrag for my wife's Vista laptop, because of its automatic defrag facility, and because it can do a complete defrag at boot time. Those annoying UAC messages interfered with the JKD screen saver. On my own laptop I use both PerfectDisk 2008 and JkDefrag, along with whatever programs I am testing at the time. I have also tried a combination of JKD and UltraDefrag's boot time defrag.

Unfortunately I know very little about C programming or I'd probably get involved in tweaking the code in either or both GPL projects.

I think Jeroen is both modest and pragmatic when he says that he would expect commercial defrag programs to work better than JKD. He has achieved a great deal with JKD, and it works incredibly well.

[Lundholm]

It turns out that measuring the effects of a defrag program is not as simple as it looks, and I'll have to go back to the drawing board to figure out another method of measuring defrag programs.

Exactly. The optimization part is a challenge, but I don't think that it is very complicated to make your benchmark testing much more realistic.

As a starting point, you can use a snapshot of any system disk with a large number of fragmented files, which has been in normal use for some time. We all have one of those.

Next step is to produce a sorted list of last access time for all files on the disk.

Last step is to build the benchmark read test to include only the most recently accessed files - maybe 50% - maybe only 33%.

The tests should include the various performance optimization options of each product under test, like JKD's sort options.

This could produce a very interesting set of tests. Can't wait to see them.

[Donn Edwards]

Lundholm I'm doing most of this already. The disk image is taken from a standard install of WinXP with all service packs and no defrag done.

00:13:02 - Total disk space: 8910364672 bytes (8.30 gigabytes), 2175382 clusters
00:13:02 - Bytes per cluster: 4096 bytes
00:13:02 - Number of files: 21733
00:13:02 - Number of directories: 2498
00:13:02 - Total size of analyzed items: 4802846720 bytes (4.47 gigabytes), 1172570 clusters
00:13:02 - Number of fragmented items: 1837, 7.58% of all items
00:13:02 - Total size of fragmented items: 1217507328 bytes, 297243 clusters, 25.35% of all items, 13.66% of disk
00:13:02 - Free disk space: 3018489856 bytes, 736936 clusters, 33.88% of disk
00:13:02 - Number of gaps: 1516
00:13:02 - Number of small gaps: 1280, 84.43% of all gaps
00:13:02 - Size of small gaps: 22507520 bytes, 5495 clusters, 0.75% of free disk space
00:13:02 - Number of big gaps: 236 (15.57% of all gaps)
00:13:02 - Size of big gaps: 2995982336 bytes, 731441 clusters, 99.25% of free disk space
00:13:02 - Average gap size: 486.11 clusters
00:13:02 - Biggest gap: 2796736512 bytes, 682797 clusters, 92.65% of free disk space

The benchmark software determines the read times of a predefined list of files. The problem is that some of the time some of the files are cached, at other times not. In order to eliminate as much of this as possible I do a reboot before reading the files.

The only difference with your suggestion is rather than using a fresh install of XP I need to run it for some time. I don't have a spare machine I can use for this purpose. The test machine is only available because it can't be used for normal use: it has a faulty cooling fan and has to be run in the fridge to keep it cool. This is fine for long tests, but not practical for normal operations.

I have obtained a test utility from Raxco that can generate fragmented files, and will experiment with this.

[Lundholm]

The only difference with your suggestion is rather than using a fresh install of XP I need to run it for some time. I don't have a spare machine I can use for this purpose.

Hi Donn,

I can understand this, but it is the key point in transforming your test from a theoretical to a realistic benchmark.

Don't you think that your local PC shop would like to sponsor these tests, and lend you an old PC? You could promise him that his name and logo will be displayed in the international test report.