Phoronix ZFS vs UFS vs EXT4 vs Btrfs

That's quite interesting! ZFS definitely seems like the winner on BSD which is surprising as you commented. As with some other filesystems, ZFS seems to focus on flexibility and features rather than all out performance. I haven't looked into Btrfs much, but does it have a similar feature set to ZFS?

Either way, competition is good in the FS arena. I don't think any of will complain if there is motivation to squeeze out more I/O on the same hardware!
 
>I don't think any of will complain if there is motivation to squeeze out more I/O on the same hardware!

Silly synthetic benchmarks aren't of any help. Comparing heavily tuned operating systems like Ubuntu with stock operating systems like FreeBSD is nonsense, comparing alpha-crap-FS like brtfs with a mature FS like ZFS too. Phoronix is an eldorado for hardware junkies with lots of time at hand.
 
The test platform is funny ... a notebook >:>
Click to expand...
To me that's what is interesting coz ZFS is performing much better than what I had expected. Though I do not have experience with ZFS the general feeling is that ZFS is a bit slower and also heavier.
 
I don't know any background on Phoronix so maybe I'm being naive, but doing this sort of test is about consistency. It shouldn't matter if you're running it on cheap or expensive hardware as long as the test conditions are the same. It tells you something about the filesystems either way. Whether it tells you something useful is a different story!

Oliverh, can you provide some background on why you say Ubuntu is "heavily tuned?" I've never seen anything that mentions extensive tuning of the default Ubuntu filesystem options.
 
Sylgeist said:
It shouldn't matter if you're running it on cheap or expensive hardware as long as the test conditions are the same.
Click to expand...
Yep, the FreeBSD results are pretty shocking if you ask me. Especially UFS, a file system that's supposed to be fast by virtue of the fact that it doesn't journal. Really hope the relevant devs look into this...
 
ZFS without proper tuning (yes it still needs tuning to get rid of the write stalls) does suck. So as said the benchmark isn't any worth.
 
Also it's worth mentioning that the testsutie that they are running(and have created from scratch) is highly optimized to run on linux using the latest and greatest gcc, while freebsd can't move beyond the 4.2 in base. And they only use stock releases, so they are not installing any ports.

So all benchmarks that rely on any "performance" is not that trustworthy as a true measure.
 
gilinko said:
Also it's worth mentioning that the testsutie that they are running(and have created from scratch) is highly optimized to run on linux using the latest and greatest gcc, while freebsd can't move beyond the 4.2 in base. And they only use stock releases, so they are not installing any ports.

So all benchmarks that rely on any "performance" is not that trustworthy as a true measure.
Click to expand...

don't know if the gcc version has that much impact on filesystem performance/benchmark.
It's true for most of the other benchmarks in their suite.
 
For me its very funny (and unprofessional at the same time) that while they want to compare FILESYSTEMS they run various RENDERING/COMPRESSION tests that do not test FILESYSTEM performance but CPU/GFX card performance ...
 
:) feel free to test your own computer >:>

Code: 
 pwd
/usr/ports/benchmarks/phoronix-test-suite

it needs php cli *g*
 
trybeingarun said:
Hey guys,

So what do you think about phoronix's comparison of various FS?
http://www.phoronix.com/scan.php?page=article&item=zfs_ext4_btrfs&num=1

Personally I do not have much to comment on ext4 and btrfs performance. To me the most surprising part was ZFS vs UFS. I thought ZFS is extremely resource consuming and will be very slow but from what Phoronix says ZFS even performs much better than UFS.

Share your thoughts.
Click to expand...

I find zfs mirror to be way faster than hardware raid1 on ufs.

my first way to test is portsnap extract, to see if I get any pauses on it. Only hardware raid10(ufs) and zfs mirror get through it with no pauses. single drive,and hw raid 1 (both on ufs) both pause.

Other tests including bonnie++, dd, and simple file copying put zfs out in front as well.

Final test server running doing normal tasks, so far zfs seems fine.

UFS seems particurly poor at sequential writes and random access.

To me UFS seems to be heading for its end of life. It seems to get little development, has poor data recovery, and is slow. What is keeping it alive until now is a lack of a competiton on freebsd.
 
Others have pointed out issues with the benchmarking which is generally correct, however there are some root issues with the linux filesystems you should be aware of. It's all fine and dandy to have a screaming fast filesystem, but if you don't have data intergrity then big whooptie-do.

btrfs - http://lkml.org/lkml/2010/6/3/313
By the time they fix all the problems with this filesystem, I suspect you'll see a dramatic loss in performance.

ex3/4 - I don't know the full testing setup, but I'll bet you dollars to doughnuts barriers weren't enabled on those file systems.
http://kernelnewbies.org/Ext4#head-25c0a1275a571f7332fa196d4437c38e79f39f63
Kind of a funny article as it states "improves the integrity of the file system" were as with out file system integrity is not guaranteed. Also note that it advises you to disable barriers to benchmark it, wtf? The final craziness to these file systems is that barriers are incompatible with device mapper, so you cannot guarantee file system integrity on things like lvm and dm-crypt partitions.
 
oliverh said:
>Silly synthetic benchmarks aren't of any help. Comparing heavily tuned operating systems like Ubuntu with stock operating systems like FreeBSD is nonsense, comparing alpha-crap-FS like brtfs with a mature FS like ZFS too. Phoronix is an eldorado for hardware junkies with lots of time at hand.
Click to expand...
I think that is too harsh. Phoronix has interesting articles; and yes benchmarking is always controversial because it can always be debated what the numbers actually mean. In other words, often the conclusions humans attach to some numbers are wrong; not the numbers themselves. This is because the number does not represent their understanding of what is being tested. Thus, when benchmarking it is very hard to make definite conclusions unless you're doing a really thorough test setup with alot of variables involved.

However, it all depends on what you want to test, what kind of conclusions you would like to make. Comparing ZFS directly against Ext4 is pretty much impossible i think. But if you're a user that won't do any tuning and will just use an OS' filesystem as-is, then benchmarks done with heavy tuning won't matter to you; you're only interested in out-of-the-box performance.

When you test Linux and FreeBSD setups for filesystem performance, what you're actually comparing is how two different Operating Systems perform with an out-of-the-box configuration under the circumstances you tested it with.

Likewise, doing tests which use I/O only partly like compression tests, can still be of interest to the user. It won't be an all-I/O benchmark so you're testing alot of variables here; but still could help answer the question whether different filesystems are actually noticeable when doing your casual tasks. If compressing files is one of them, such a test can still be worthwhile.

Also keep in mind Phoronix isn't commercial (or at least doesn't appear to be). If you think they could be doing the tests better, then help them with constructive criticism and turn it into advise rather than discrediting a nice hobby website. Also keep in mind that benchmarking properly is extremely hard. I've not seen a single review yet where i think it comes close to a perfect test setup.

My thoughts.
-sub
 
You must log in or register to reply here.
Back
Top