Performance figures with and without relocated journal

Performance figures with and without relocated journal

[Philippe Gramoullé]

 Hi,

 I would like to know if there are some figures available for filesystem benchmarks
some done without a relocated journal and some done with a relocated one 
( even better if tests with NFS are available )

 What order of performance improvement can i hope to get with a relocated journal ?

 Thanks,

Philippe

Re: Performance figures with and without relocated journal

[Edward Shushkin]

Philippe Gramoullé wrote:
> 
>  Hi,
> 
>  I would like to know if there are some figures available for filesystem benchmarks
> some done without a relocated journal and some done with a relocated one
> ( even better if tests with NFS are available )
> 
>  What order of performance improvement can i hope to get with a relocated journal ?

We still don't have comprehensive results in this area,
but I should note there is some improvement for data logging journal mode:
http://namesys.com/benchmarks/journal_relocation_to_NVRAM.html

Edward.


> 
>  Thanks,
> 
> Philippe

reiserfs vs NTFS vs WAFL

[darren]

Hi all,

I wonder if there are any papers or benchmarks that compare the
performance and architectural differences between reiserfs, NTFS and
WAFL (NetApp's Write Anywhere File Layout)?

Any help will be great.

Regards
darren

Re: reiserfs vs NTFS vs WAFL

[Hans Reiser]

darren wrote:

>Hi all,
>
>I wonder if there are any papers or benchmarks that compare the
>performance and architectural differences between reiserfs, NTFS and
>WAFL (NetApp's Write Anywhere File Layout)?
>
>Any help will be great.
>
>Regards
>darren
>
>
>
>  
>
I regret to say that all I can do is point you to my LinuxTag paper at 
www.namesys.com/v4/reiser4_the_atomic_filesystem.html, and give you my 
informal opinion that NTFS sucks but has improved a lot, and WAFL is 
great for NFS.

I'd be curious to see mongo numbers on NTFS if anyone has the energy.

Most people comparing WAFL and ReiserFS look at the price per GB and we 
win without the benchmark ever being run.  mp3.com for instance 
sponsored us so that they could save $22 million on Netapps and Suns. 

I would expect WAFL to beat Linux/NFS/ReiserFS V3/RAID because that is 
their niche and they are sharp and I bet their zero copy NFS code is 
better than Linux's, but no one knows for real.  I would be curious to 
see Reiser4 benchmarked against WAFL when we are done, especially if you 
toss in some cards from umem that Mike sells.  ReiserFS might save some 
disk space compared to WAFL.

I would expect NetApp to just blow away any NTFS server.... but I 
haven't measured it.

-- 
Hans

Re: reiserfs vs NTFS vs WAFL

[Ragnar Kjørstad]

On Wed, May 21, 2003 at 10:13:57PM +0800, darren wrote:
> Hi all,
> 
> I wonder if there are any papers or benchmarks that compare the
> performance and architectural differences between reiserfs, NTFS and
> WAFL (NetApp's Write Anywhere File Layout)?

I did some benchmarking of linux vs NetApp benchmarking a long time
ago.

We used "postmark" to perform the benchmark, with default settings,
except:
number 20000
transactions 50000

Information on postmark is available at
http://www.netapp.com/tech_library/3022.html. Later I have learned that
postmark is probably not the best general-purpose benchmarking tool, so
the numbers may not be representative for the average workload.

All tests were done with a single NFS-client, a linux PC. (linux 2.2,
NFS v2).

The servers were:
1. NetApp 760, with FC disks
2. x86 PC, single SCSI-disk, linux 2.2, ext2, NFSv2
3. x86 PC, single SCSI-disk, linux 2.2, reiserfs v3.5.5, NFSv2


Results, transactions pr second:
1: 82 transactions/s
2: 43 transactions/s
3: 66 transactions/s

So, in this test the $100k NetApp was approxemately 24% faster than the
$1k linux workstation.


Now, there are many obvious flaws in this test, in no particular order:
- The benchmark-tool may not be the best
- A single client is not sufficient to represent a typical production-
  enviroment
- The linux-server should have had RAID
- The tests should have been repeated and averaged
- It's old. Terribly old.


So, take the results with a grain of salt.
It would be great run some new benchmarks (e.g. SPEC sfs) on a properly
spec'ed linux/reiserfs system, but personally I don't have the hardware
available just now.



-- 
Ragnar Kjørstad
Zet.no

RE: reiserfs vs NTFS vs WAFL

[darren]

Hi all,

Thanx for replying to my mail.

I am doing some tests on this three FS now (to see if we would want to
spend $$$$$ on a netapp).

My test is based on the simultaneous creation and deletion of small
files 50Kb, as per my actual usage of the file server for my application
servers.

My main problem with using Linux/Reiserfs3 is with the performance of
Samba when under heavy load. CIFS operation is required as some of our
application servers are windows based. Netapp's CIFS performance is
really incredible.

With NFS, we can sometimes beat the netapp F87 with a
Linux/ReiserFs/RAID0 config with the same number of harddisk.

Will post more details when I thru with it.

In the mean time, will really appreciate any suggestions to improving
Samba's performance in order to make a CIFS file server based on
Linux/Reiser/Samba possible.

-----Original Message-----
From: Hans Reiser [mailto:reiser@namesys.com] 
Sent: Thursday, May 22, 2003 2:14 AM
To: darren
Cc: 'reiserfs'; Mike Jadon
Subject: Re: reiserfs vs NTFS vs WAFL

darren wrote:

>Hi all,
>
>I wonder if there are any papers or benchmarks that compare the
>performance and architectural differences between reiserfs, NTFS and
>WAFL (NetApp's Write Anywhere File Layout)?
>
>Any help will be great.
>
>Regards
>darren
>
>
>
>  
>
I regret to say that all I can do is point you to my LinuxTag paper at 
www.namesys.com/v4/reiser4_the_atomic_filesystem.html, and give you my 
informal opinion that NTFS sucks but has improved a lot, and WAFL is 
great for NFS.

I'd be curious to see mongo numbers on NTFS if anyone has the energy.

Most people comparing WAFL and ReiserFS look at the price per GB and we 
win without the benchmark ever being run.  mp3.com for instance 
sponsored us so that they could save $22 million on Netapps and Suns. 

I would expect WAFL to beat Linux/NFS/ReiserFS V3/RAID because that is 
their niche and they are sharp and I bet their zero copy NFS code is 
better than Linux's, but no one knows for real.  I would be curious to 
see Reiser4 benchmarked against WAFL when we are done, especially if you

toss in some cards from umem that Mike sells.  ReiserFS might save some 
disk space compared to WAFL.

I would expect NetApp to just blow away any NTFS server.... but I 
haven't measured it.

-- 
Hans

Re: reiserfs vs NTFS vs WAFL

[Hans Reiser]

darren wrote:

>Hi all,
>
>Thanx for replying to my mail.
>
>I am doing some tests on this three FS now (to see if we would want to
>spend $$$$$ on a netapp).
>
>My test is based on the simultaneous creation and deletion of small
>files 50Kb, as per my actual usage of the file server for my application
>servers.
>
>My main problem with using Linux/Reiserfs3 is with the performance of
>Samba when under heavy load. CIFS operation is required as some of our
>application servers are windows based. Netapp's CIFS performance is
>really incredible.
>
>With NFS, we can sometimes beat the netapp F87 with a
>Linux/ReiserFs/RAID0 config with the same number of harddisk.
>
This really surprises me.  Be sure to turn tails off when doing a purely 
performance benchmark with V3.

>
>Will post more details when I thru with it.
>
>In the mean time, will really appreciate any suggestions to improving
>Samba's performance in order to make a CIFS file server based on
>Linux/Reiser/Samba possible.
>
Sponsor a guy to work on it.;-)  Sorry, I know that you are probably 
much smaller than mp3.com and can't really cost justify that.  For a 
long time we have known that it would be nice to have someone tweak 
samba for reiserfs, but there is no funding for it (which is odd 
considering the practical/commercial import of such work).

>
>-----Original Message-----
>From: Hans Reiser [mailto:reiser@namesys.com] 
>Sent: Thursday, May 22, 2003 2:14 AM
>To: darren
>Cc: 'reiserfs'; Mike Jadon
>Subject: Re: reiserfs vs NTFS vs WAFL
>
>darren wrote:
>
>  
>
>>Hi all,
>>
>>I wonder if there are any papers or benchmarks that compare the
>>performance and architectural differences between reiserfs, NTFS and
>>WAFL (NetApp's Write Anywhere File Layout)?
>>
>>Any help will be great.
>>
>>Regards
>>darren
>>
>>
>>
>> 
>>
>>    
>>
>I regret to say that all I can do is point you to my LinuxTag paper at 
>www.namesys.com/v4/reiser4_the_atomic_filesystem.html, and give you my 
>informal opinion that NTFS sucks but has improved a lot, and WAFL is 
>great for NFS.
>
>I'd be curious to see mongo numbers on NTFS if anyone has the energy.
>
>Most people comparing WAFL and ReiserFS look at the price per GB and we 
>win without the benchmark ever being run.  mp3.com for instance 
>sponsored us so that they could save $22 million on Netapps and Suns. 
>
>I would expect WAFL to beat Linux/NFS/ReiserFS V3/RAID because that is 
>their niche and they are sharp and I bet their zero copy NFS code is 
>better than Linux's, but no one knows for real.  I would be curious to 
>see Reiser4 benchmarked against WAFL when we are done, especially if you
>
>toss in some cards from umem that Mike sells.  ReiserFS might save some 
>disk space compared to WAFL.
>
>I would expect NetApp to just blow away any NTFS server.... but I 
>haven't measured it.
>
>  
>


-- 
Hans

Re: reiserfs vs NTFS vs WAFL

[Ragnar Kjørstad]

On Thu, May 22, 2003 at 10:39:56PM +0800, darren wrote:
> Hi all,
> 
> Thanx for replying to my mail.
> 
> I am doing some tests on this three FS now (to see if we would want to
> spend $$$$$ on a netapp).

Be sure to post whatever you find here. I'm sure I'm not the only
curious one.

> My test is based on the simultaneous creation and deletion of small
> files 50Kb, as per my actual usage of the file server for my application
> servers.

There is a small frontend-script to bonnie++ available at
http://www.stud.ntnu.no/~ragnarkj/download/dbonnie++/dbonnie++.py if
you're interested.

It aims to assist in running bonnie++ at multiple clients at once.

It's very old, so it may or may not work with recent versions of
bonnie++.

Alternatively SPEC sfs is probably a good way to test NFS-performance
(but it costs $$).

> My main problem with using Linux/Reiserfs3 is with the performance of
> Samba when under heavy load. CIFS operation is required as some of our
> application servers are windows based. Netapp's CIFS performance is
> really incredible.

That is realy surprising. 
Samba is just another userspace application. I can't see how reiserfs
could perform worse with samba than with other applications? 

Did you test other filesystems as well? e.g. xfs?

It would be really interesting to know more about what the problem is. 


> With NFS, we can sometimes beat the netapp F87 with a
> Linux/ReiserFs/RAID0 config with the same number of harddisk.

I take it that means it's slower with RAID5?
Even so, you can probably get a linux-system with RAID10 (and twice the
number of disks) for the price of a netapp...

> Will post more details when I thru with it.
Please do.

> In the mean time, will really appreciate any suggestions to improving
> Samba's performance in order to make a CIFS file server based on
> Linux/Reiser/Samba possible.

There are some generic samba-tuning information available at:
http://hr.uoregon.edu/davidrl/samba/samba-optimize.html


I would also suggest trying to figgure out exactly what it is that is
slow with samba on reiserfs. Are there specific operations that are very
expensive? Is the server network-, cpu- or disk-bound?
Things like that.



-- 
Ragnar Kjørstad
Zet.no

Re: reiserfs vs NTFS vs WAFL

[Hans Reiser]

Ragnar Kjørstad wrote:

>
>Alternatively SPEC sfs is probably a good way to test NFS-performance
>(but it costs $$).
>
I am told it is free now.


-- 
Hans

Re: reiserfs vs NTFS vs WAFL

[Ragnar Kjørstad]

On Thu, May 22, 2003 at 09:22:23PM +0400, Hans Reiser wrote:
> Ragnar Kjørstad wrote:
> 
> >Alternatively SPEC sfs is probably a good way to test NFS-performance
> >(but it costs $$).
> >
> I am told it is free now.

If that's true I might give it ago myself, but according to
http://www.spec.org/cgi-bin/order it is 900$.

That's a little too expensive for me just beeing curious.


-- 
Ragnar Kjørstad
Zet.no

RE: reiserfs vs NTFS vs WAFL

[darren]

Hi Hans,

> This really surprises me.  Be sure to turn tails off when doing a
purely 
performance benchmark with V3.

I mounted the drive with the notails option during the test already.

> Sponsor a guy to work on it.;-)  Sorry, I know that you are probably 
much smaller than mp3.com and can't really cost justify that.  For a 
long time we have known that it would be nice to have someone tweak 
samba for reiserfs, but there is no funding for it (which is odd 
considering the practical/commercial import of such work).

You are so right!!!...i am just a student working on a school project!!!
While we have some funding...it is so limited we have to thrift on RAM,
and warranty for our servers....sigh...

But seriously, CIFS have a lot of commercial appeal now considering the
number of people who already bought NetApps to run their Databases on
windows.

Even running an enterprise or campus Fileserver, we cannot ignore the
people still in the matrix and running windoze.

Thanx for reiserfs really!


Rgds
darren

-----Original Message-----
From: Hans Reiser [mailto:reiser@namesys.com] 
Sent: Thursday, May 22, 2003 11:25 PM
To: darren
Cc: 'Ragnar Kjorstad'; 'reiserfs'; 'Mike Jadon'
Subject: Re: reiserfs vs NTFS vs WAFL

darren wrote:

>Hi all,
>
>Thanx for replying to my mail.
>
>I am doing some tests on this three FS now (to see if we would want to
>spend $$$$$ on a netapp).
>
>My test is based on the simultaneous creation and deletion of small
>files 50Kb, as per my actual usage of the file server for my
application
>servers.
>
>My main problem with using Linux/Reiserfs3 is with the performance of
>Samba when under heavy load. CIFS operation is required as some of our
>application servers are windows based. Netapp's CIFS performance is
>really incredible.
>
>With NFS, we can sometimes beat the netapp F87 with a
>Linux/ReiserFs/RAID0 config with the same number of harddisk.
>
This really surprises me.  Be sure to turn tails off when doing a purely

performance benchmark with V3.

>
>Will post more details when I thru with it.
>
>In the mean time, will really appreciate any suggestions to improving
>Samba's performance in order to make a CIFS file server based on
>Linux/Reiser/Samba possible.
>
Sponsor a guy to work on it.;-)  Sorry, I know that you are probably 
much smaller than mp3.com and can't really cost justify that.  For a 
long time we have known that it would be nice to have someone tweak 
samba for reiserfs, but there is no funding for it (which is odd 
considering the practical/commercial import of such work).

>
>-----Original Message-----
>From: Hans Reiser [mailto:reiser@namesys.com] 
>Sent: Thursday, May 22, 2003 2:14 AM
>To: darren
>Cc: 'reiserfs'; Mike Jadon
>Subject: Re: reiserfs vs NTFS vs WAFL
>
>darren wrote:
>
>  
>
>>Hi all,
>>
>>I wonder if there are any papers or benchmarks that compare the
>>performance and architectural differences between reiserfs, NTFS and
>>WAFL (NetApp's Write Anywhere File Layout)?
>>
>>Any help will be great.
>>
>>Regards
>>darren
>>
>>
>>
>> 
>>
>>    
>>
>I regret to say that all I can do is point you to my LinuxTag paper at 
>www.namesys.com/v4/reiser4_the_atomic_filesystem.html, and give you my 
>informal opinion that NTFS sucks but has improved a lot, and WAFL is 
>great for NFS.
>
>I'd be curious to see mongo numbers on NTFS if anyone has the energy.
>
>Most people comparing WAFL and ReiserFS look at the price per GB and we

>win without the benchmark ever being run.  mp3.com for instance 
>sponsored us so that they could save $22 million on Netapps and Suns. 
>
>I would expect WAFL to beat Linux/NFS/ReiserFS V3/RAID because that is 
>their niche and they are sharp and I bet their zero copy NFS code is 
>better than Linux's, but no one knows for real.  I would be curious to 
>see Reiser4 benchmarked against WAFL when we are done, especially if
you
>
>toss in some cards from umem that Mike sells.  ReiserFS might save some

>disk space compared to WAFL.
>
>I would expect NetApp to just blow away any NTFS server.... but I 
>haven't measured it.
>
>  
>


-- 
Hans

RE: reiserfs vs NTFS vs WAFL

[darren]

Hi,

I don't think the performance lies with Reiserfs per say, but if I want
to make full use of the Reiserfs to host a fileserver for windows
clients, Samba would be my bottleneck.

I noticed that my samba server takes up a huge chuck of my CPU after a
long period of file writes into the fileserver.

Have already applied whatever optimization to samba that I can find but
no help.

Its incredible how the NetApp's CIFS implementation can work so fast...


-----Original Message-----
From: Ragnar Kjørstad [mailto:reiserfs@ragnark.vestdata.no] 
Sent: Friday, May 23, 2003 1:07 AM
To: darren
Cc: 'Hans Reiser'; 'reiserfs'; 'Mike Jadon'
Subject: Re: reiserfs vs NTFS vs WAFL

On Thu, May 22, 2003 at 10:39:56PM +0800, darren wrote:
> Hi all,
> 
> Thanx for replying to my mail.
> 
> I am doing some tests on this three FS now (to see if we would want to
> spend $$$$$ on a netapp).

Be sure to post whatever you find here. I'm sure I'm not the only
curious one.

> My test is based on the simultaneous creation and deletion of small
> files 50Kb, as per my actual usage of the file server for my
application
> servers.

There is a small frontend-script to bonnie++ available at
http://www.stud.ntnu.no/~ragnarkj/download/dbonnie++/dbonnie++.py if
you're interested.

It aims to assist in running bonnie++ at multiple clients at once.

It's very old, so it may or may not work with recent versions of
bonnie++.

Alternatively SPEC sfs is probably a good way to test NFS-performance
(but it costs $$).

> My main problem with using Linux/Reiserfs3 is with the performance of
> Samba when under heavy load. CIFS operation is required as some of our
> application servers are windows based. Netapp's CIFS performance is
> really incredible.

That is realy surprising. 
Samba is just another userspace application. I can't see how reiserfs
could perform worse with samba than with other applications? 

Did you test other filesystems as well? e.g. xfs?

It would be really interesting to know more about what the problem is. 


> With NFS, we can sometimes beat the netapp F87 with a
> Linux/ReiserFs/RAID0 config with the same number of harddisk.

I take it that means it's slower with RAID5?
Even so, you can probably get a linux-system with RAID10 (and twice the
number of disks) for the price of a netapp...

> Will post more details when I thru with it.
Please do.

> In the mean time, will really appreciate any suggestions to improving
> Samba's performance in order to make a CIFS file server based on
> Linux/Reiser/Samba possible.

There are some generic samba-tuning information available at:
http://hr.uoregon.edu/davidrl/samba/samba-optimize.html


I would also suggest trying to figgure out exactly what it is that is
slow with samba on reiserfs. Are there specific operations that are very
expensive? Is the server network-, cpu- or disk-bound?
Things like that.



-- 
Ragnar Kjørstad
Zet.no

Re: reiserfs vs NTFS vs WAFL

[Heinz-Josef Claes]

[-- Attachment #1: Type: text/plain, Size: 1000 bytes --]

Am Mit, 2003-05-21 um 16.13 schrieb darren:
> Hi all,
> 
> I wonder if there are any papers or benchmarks that compare the
> performance and architectural differences between reiserfs, NTFS and
> WAFL (NetApp's Write Anywhere File Layout)?
> 
> Any help will be great.
> 
> Regards
> darren

I made some tests with nfs about three years ago with linux and a filer.
All I remember is that it was a smaller machine from NetApp. 
All documentation I still have from the test is a sheet, in which I
directly typed the times :-(
It's appended to this mail; sorry for making an attachement (OpenOffice
format), but it's smaller than some ordinary mail in this list.

Hope, this helps a little bit.

-- 
Heinz-Josef Claes                     hjclaes@web.de
project: http://sourceforge.net/projects/storebackup
      -> snapshot-like backup to another disk

Q: How does a Unix guru have sex?
A: gunzip && strip && touch && finger && mount && fsck && \
more && yes && fsck && fsck && fsck && umount && sleep


[-- Attachment #2: testFiler.sxc --]
[-- Type: application/vnd.sun.xml.calc, Size: 7936 bytes --]

Re: reiserfs vs NTFS vs WAFL

[Hans Reiser]

Heinz-Josef Claes wrote:

>Am Mit, 2003-05-21 um 16.13 schrieb darren:
>  
>
>>Hi all,
>>
>>I wonder if there are any papers or benchmarks that compare the
>>performance and architectural differences between reiserfs, NTFS and
>>WAFL (NetApp's Write Anywhere File Layout)?
>>
>>Any help will be great.
>>
>>Regards
>>darren
>>    
>>
>
>I made some tests with nfs about three years ago with linux and a filer.
>All I remember is that it was a smaller machine from NetApp. 
>All documentation I still have from the test is a sheet, in which I
>directly typed the times :-(
>It's appended to this mail; sorry for making an attachement (OpenOffice
>format), but it's smaller than some ordinary mail in this list.
>
>Hope, this helps a little bit.
>
>  
>
This suggests that Reiserfs does writes faster over NFS than locally?

-- 
Hans