I've been wondering how much of a performance hit a journaled file system would take compared to a non-journaled one. Rumors had it that it was quite considerable, but was going to be much less in Mac OS X v10.3 aka Panther.
When I was comparing PCI vs OS X RAID in Mac OS 10.2.8 I did some preliminary tests with journaling as well. Results were similar to these later and more extensive tests in Panther, hence no 10.2.8 results are included.
This time, I didn't do any tests with benchmarking tools at all.
Test setup:
PowerMac G4, dual 1.42 GHz, 2 GB RAM, Radeon 9800 Pro, Mac OS X v10.3 (Build 7B85)
Two WD 200 GB 7200 rpm 8 MB cache drives, each attached as Master to an individual channel on an Acard ATA-133 PCI RAID card.
Usually one HD was formatted with "Journaled HFS Extended" while the other one was formatted with the regular "HFS extended" for test1, and vice versa for test2.
file: a single 1.36 GB file (ASR disk image), folder: a 351.2 MB folder containing 32'692 items
Finder file/folder duplication
In this test, I simply duplicated a large file (i) and a large folder (ii). As I couldn't believe the folder duplication results at first, I made another test (iii) with the two folders I had after the second test (rebooted the machine first, to avoid caching effects).
All of the tests were done twice (on different harddisks) and averaged, just to be sure.
| journaled | test 1 | test 2 | average | non-journaled | test 1 | test 2 | average |
| i. file duplicate | 67 | 64 | 65.5 | i. file duplicate | 64 | 65 | 64.5 |
| ii. folder duplicate | 113 | 114 | 113.5 | ii. folder duplicate | 124 | 124 | 124 |
| iii. 2 folder duplicate | 237 | 233 | 235 | iii. 2 folder duplicate | 256 | 253 | 254.5 |
As expected, duplicating a single large file is pretty much the same speed for journaled/non-journaled HFS (it's a very small journaling overhead compared to long time for actual duplication).
Very suprising (at least for me) were the results of the folder duplication, because it was *faster* when using the journaled filesystem! That's why I did the test again with 2 of the folders, with the same results. I would have expected the journaled HFS to be slower due to the journaling overhead for each of the many items.
Finder copy/duplicate/copy tests
In this test I copied a file/folder to another harddrive, duplicated it on the second harddrive, and copied the duplicate back to the original harddrive. Both harddrives used either journaled or non-journaled HFS Extended.
| journaled | test 1 | test 2 | average | non-journaled | test 1 | test 2 | average |
| copy folder A to B | 88 | 90 | 89 | copy folder A t | 89 | 91 | 90 |
| duplicate on B | 108 | 109 | 108.5 | duplicate on B | 114 | 118 | 116 |
| copy back B to A | 93 | 92 | 92.5 | copy back B to A | 97 | 95 | 96 |
| Total Time | 289 | 291 | 290 | 300 | 304 | 302 | |
| copy file A to B | 45 | 46 | 45.5 | copy file A to | 46 | 46 | 46 |
| duplicate on B | 65 | 63 | 64 | duplicate on B | 63 | 64 | 63.5 |
| copy back B to A | 47 | 46 | 46.5 | copy back B to | 46 | 46 | 46 |
| Total Time | 157 | 155 | 156 | 155 | 156 | 155.5 |
The file copy/duplicate/copy sequence was pretty much equally fast for both journaled/non-journaled HFS.
The same sequence with the folder was again somewhat faster with the journaled HFS, although the largest time difference was in the duplication part, and not the copying.
Cheers, Oliver aka boli
Related Links:
xlr8yourmac Forum thread: Benefits of a Journaling File System
xlr8yourmac Forum thread: A guess as to why things can be faster with journaling
xlr8yourmac Forum thread: More about journaling (and OS 9)
ArsTechnica Forum thread: I asked why I got the results above (only one reply, though)