Comparing different file systems
Hi list! I have run a couple of benchmarks to figure out how the file system affects your boot up speed. I found that by simply installing Suse 10.0 on Ext3 instead of ReiserFS, the boot time went down from 81 seconds to just 58! Please have a look at the article, tell me what you think about this and add your own measurements if you have any. The article is available at http://www.opensuse.org/File_systems, so you can add your own data directly. Cheers nordi
Sunday 09 Oct 2005 19:24 samaye nordi alekhiit:
affects your boot up speed. I found that by simply installing Suse 10.0 on Ext3 instead of ReiserFS, the boot time went down from 81 seconds to just 58!
Just a question of curiosity - why would that be?
Shriramana Sharma wrote:
Just a question of curiosity - why would that be?
Good question! I also wondered why the difference was _that_ big. Here are my thoughts, copied from the website: "My personal guess is that this is because ReiserFS needs lots of CPU time (at least that's what I heard). This doesn't matter in your normal "copy lots of files" benchmark, since in those scenarios the CPU is almost idle. However, when you boot a system, the CPU is not idle but has lots of stuff to do, which might slow down a CPU intensive file system. Note however, that this is just my personal guess." Cheers nordi
Sunday 09 Oct 2005 23:44 samaye nordi alekhiit:
Just a question of curiosity - why would that be?
Good question! I also wondered why the difference was _that_ big. Here are my thoughts, copied from the website:
Yeah I read that, but:
"My personal guess is that this is because ReiserFS needs lots of CPU time (at least that's what I heard).
... the question then becomes - *why* would reiserfs need lots of CPU? What is it inside reiserfs that makes it so? S:R
Hello,
On 10/10/05, Shriramana Sharma
Yeah I read that, but:
... the question then becomes - *why* would reiserfs need lots of CPU? What is it inside reiserfs that makes it so?
There is something strange with reiserfs. One year ago, when I tried Gentoo reiserfs was flying in comparison with ext3. I also did some tests on Slackware 10.0 using ext2, ext3, reiserfs, xfs and reiserfs was the fastest one. Now I see the tests on SuSE and I want to go back to ext3 because speed is very important for me. Yours faithfully, -- Damian Mihai Liviu Phone: +40741226993 Yahoo: liviudm_cisco URL: http://liviudm.blogspot.com
Monday 10 Oct 2005 11:32 samaye Dazzle alekhiit:
was the fastest one. Now I see the tests on SuSE and I want to go back to ext3 because speed is very important for me.
Well you know what they say - your mileage may vary. Nordi's test meant that for him ext3 was faster. Now maybe I've missed some mails in this thread but until someone else has confirmed Nordi's on a widely different configuration of a PC, one cannot conclude that the problem is with reiserfs. So don't jump out of the window yet...
Shriramana Sharma wrote:
Well you know what they say - your mileage may vary. Nordi's test meant that for him ext3 was faster. Now maybe I've missed some mails in this thread but until someone else has confirmed Nordi's on a widely different configuration of a PC, one cannot conclude that the problem is with reiserfs. Exactly! That's also the reason why I'd like other people to make benchmarks of unmodified ReiserFS vs unmodified Ext3. The only other PC I have available for testing is a K6-2 300Mhz, and that would certainly not be representative for your average Suse PC.
Cheers nordi
On Monday 10 October 2005 3:31 am, nordi wrote:
Shriramana Sharma wrote:
Well you know what they say - your mileage may vary. Nordi's test meant that for him ext3 was faster. Now maybe I've missed some mails in this thread but until someone else has confirmed Nordi's on a widely different configuration of a PC, one cannot conclude that the problem is with reiserfs.
Exactly! That's also the reason why I'd like other people to make benchmarks of unmodified ReiserFS vs unmodified Ext3. The only other PC I have available for testing is a K6-2 300Mhz, and that would certainly not be representative for your average Suse PC.
I have a pentium III 600 i'll test it on. Again, not a 'typical' pc, but at least it's something. Kirk
On Monday 10 October 2005 17:14, Kirk Coombs wrote:
On Monday 10 October 2005 3:31 am, nordi wrote:
Shriramana Sharma wrote:
Well you know what they say - your mileage may vary. Nordi's test meant that for him ext3 was faster. Now maybe I've missed some mails in this thread but until someone else has confirmed Nordi's on a widely different configuration of a PC, one cannot conclude that the problem is with reiserfs.
Exactly! That's also the reason why I'd like other people to make benchmarks of unmodified ReiserFS vs unmodified Ext3. The only other PC I have available for testing is a K6-2 300Mhz, and that would certainly not be representative for your average Suse PC.
I have a pentium III 600 i'll test it on. Again, not a 'typical' pc, but at least it's something.
I have did a quick test of it, and I can't confirm it. I was measuring a bit different numbers -- a time till login window appears (which is what probably matters for me), I have got the same numbers for both ext3 and reiser. Are you sure your results are not due to different hdd read speed for different partitions? -- Best regards, Alexander.
Alexander S. Usov wrote:
Are you sure your results are not due to different hdd read speed for different partitions?
It was all installed on the same partition, so that cannot be the reason. I also reinstalled with ReiserFS, just to make sure it was not just something really weird. But the boot time stayed the same, much slower than ext3. Given that results on your systems did not show big differences between ReiserFS and Ext3, can you give me your system configuration (processor, RAM, laptop/desktop) and also tell me what the size of your Suse partition is? Cheers nordi
On Tuesday 11 October 2005 01:58, nordi wrote:
Alexander S. Usov wrote:
Are you sure your results are not due to different hdd read speed for different partitions?
It was all installed on the same partition, so that cannot be the reason. I also reinstalled with ReiserFS, just to make sure it was not just something really weird. But the boot time stayed the same, much slower than ext3.
Given that results on your systems did not show big differences between ReiserFS and Ext3, can you give me your system configuration (processor, RAM, laptop/desktop) and also tell me what the size of your Suse partition is?
Desktop system, 2GHz P4, 512Mb ram. Root partition is approx 6 GB big, and home is on separate partition. -- Best regards, Alexander.
On Monday 10 October 2005 9:14 am, Kirk Coombs wrote:
I have a pentium III 600 i'll test it on. Again, not a 'typical' pc, but at least it's something.
Okay, here are my results. I tried to duplicate nordi's situation -- autologin and get the uptime after KDE starts. I also ran an external stopwatch from GRUB to KDE start (when the uptime window appeared). Obviously the stopwatch is prone to human error. I had a single swap and root partition, both on the same drive. Reiser: uptime: 68.28 (68.78,68.39,70.37,65.59) stopwatch: 77.11 (74.718,79.850,77.261,76.606) Ext3: uptime: 62.905 (64.65,63.36,60.16,63.45) stopwatch: 62.86 (63,61.883,62.820,63.766) Conclusion: If basing the count on uptime, ext3 is 7.8% faster. If basing it on total boot time, it is 18.4% faster. When does the uptime count start? I would have thought it would be before mounting the filesystems. It seems that it must be after, and it takes longer to mount Reiser leading to a larger gap between uptime in stopwatch time. Given that they are not mounted/unmounted frequently during use of the computer this discrepancy is probably negligible. Kirk
Kirk, On Monday 10 October 2005 12:08, Kirk Coombs wrote:
On Monday 10 October 2005 9:14 am, Kirk Coombs wrote:
I have a pentium III 600 i'll test it on. Again, not a 'typical' pc, but at least it's something.
Okay, here are my results.
I tried to duplicate nordi's situation -- autologin and get the uptime after KDE starts. I also ran an external stopwatch from GRUB to KDE start (when the uptime window appeared). Obviously the stopwatch is prone to human error.
I had a single swap and root partition, both on the same drive.
Reiser: uptime: 68.28 (68.78,68.39,70.37,65.59) stopwatch: 77.11 (74.718,79.850,77.261,76.606)
Ext3: uptime: 62.905 (64.65,63.36,60.16,63.45) stopwatch: 62.86 (63,61.883,62.820,63.766)
As with any "benchmark," it only tells you about how long it takes to run the benchmark. Everything else is an extrapolation, and probably an unwarranted one. In all likelihood, the activities during system start-up are not characteristic of those performed during normal operations. Measure the things you do routinely or that are bottlenecks in your ordinary, daily operations. If that's rebooting, then fine, but what's the typical user's ratio of reboots to days? For me it's a small number (in fact, one I try to make as small as possible). It probably does not often exceed one.
Conclusion: If basing the count on uptime, ext3 is 7.8% faster. If basing it on total boot time, it is 18.4% faster.
Conclusion: Booting on your system is from 7.8% to 18.4% faster as things stand at the moment. Any further conclusion is baseless and invalid. You just don't have the data to justify more.
...
Kirk
Randall Schulz
Randall R Schulz wrote:
In all likelihood, the activities during system start-up are not characteristic of those performed during normal operations.
True, but if the mere usage of ReiserFS slows down things by 23 seconds on my box, don't you think there is _something_ wrong here? That is a 40% difference! And assuming that the system also does lots of non-fs stuff during boot, the performance difference is probably even higher. Maybe ReiserFS is not a good choice for small partitions (my 4GB is rather small, nowaday), or it is not a good choice on laptops, or Suse sets something up incorrectly, or whatever. I think it would be really interesting to find the answer to this question. Regards nordi
Nordi, On Tuesday 11 October 2005 01:59, nordi wrote:
Randall R Schulz wrote:
In all likelihood, the activities during system start-up are not characteristic of those performed during normal operations.
True, but if the mere usage of ReiserFS slows down things by 23 seconds on my box, don't you think there is _something_ wrong here? That is a 40% difference! And assuming that the system also does lots of non-fs stuff during boot, the performance difference is probably even higher.
No. That conclusion _does not follow_ from the facts you have. It's not just a matter of "how much file system I/O" vs. "how much other stuff" it does. It's the nature of the activity, the size and placement of the files, the size of the I/O operations and whether access is largely continuous or not, whether it is aligned to page boundaries, the opportunity (or lack thereof) for concurrency in the code that drives the FS activity, etc. Very few people understand the complexity of performance analysis in modern computing systems. To think that one simple, uncontrolled experiment can tell you something about the inherent efficiency of a particular file system or implementation thereof or that it can be extrapolated to other contexts is simply incorrect. Is boot time the be-all and end-all of your system performance criteria? If so, then by all means, make whatever choices optimize it. But it's likely you cannot simultaneously optimize both start-up and normal operations.
Maybe ReiserFS is not a good choice for small partitions (my 4GB is rather small, nowaday), or it is not a good choice on laptops, or Suse sets something up incorrectly, or whatever. I think it would be really interesting to find the answer to this question.
Yes, it would. Especially if you include the null hypothesis "nothing is wrong." You're approaching this whole thing with an assumption that may or may not be the case.
Regards nordi
Randall Schulz
On Tuesday 11 October 2005 18:04, Randall R Schulz wrote:
Very few people understand the complexity of performance analysis in modern computing systems. To think that one simple, uncontrolled
That is very true. However, it would be stupid to overlook something that might actually indicate that there's something wrong before analyzing it. -- // Janne
participants (7)
-
Alexander S. Usov -
Dazzle -
Janne Karhunen -
Kirk Coombs -
nordi -
Randall R Schulz -
Shriramana Sharma