Re: sysbench / fileio - Linux vs. FreeBSD

2010-06-06 Thread Adam PAPAI
-BEGIN PGP SIGNED MESSAGE-
Hash: SHA1

On 6/5/10 2:43 AM, Igor Mozolevsky wrote:
> On 5 June 2010 00:58, Adam PAPAI  wrote:
> 
>> How can I tune my disk to make it faster? Is it possible? What is the
>> reason of the really slow I/O with more than 4 threads? What do you
>> recommend me to do? Why is it damn slow with 8K blocksize?
> 
> Does linux still have async disk writes by default?

Anyway, I looked after the default ext3 values:

Debian mounts the ext3 with "defaults" option.

This means: rw, suid, dev, exec, auto, nouser, and async.

Well it means I have to test it with UFS (async) and Debian (sync).

These test will take some time but I hope it worth the effort.

Hm...


- -- 
Adam PAPAI
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.10 (Darwin)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/

iQEcBAEBAgAGBQJMC31LAAoJEGq0EWvh5uiIu3MH/i7KWfcYj2zXSsqbUK2W4dKi
B0+pD861FBtxmS+O4c4jzR5vJYeVVyVfZ4DLpHs0tqr6u2QZWgTD5c9GXxRNn9Hg
pVIL8/iL9BGtjNZdbjKU2RlE+QOb4LUuxqTWtz3poH4e6CQlAMOzvBcmbK41eWVn
nr2/jlS8n7TFk74ewAH9NXABrhIaOtCjBf5YWWA9AnKhqjdlAM7gxC6QcbsGTLlR
5zvq6UfGuAMECOV98FDlm3k20LydLT0/Mdw9jth9+50v1NMnAddYjfZ/7Ci2KzZo
uUN1VRcOhxmw6oliMPu/+Z324d6Xrp1vXpDQN8tSzME1d3O3CswPDfs3ocpjmkU=
=VEsH
-END PGP SIGNATURE-
___
freebsd-questions@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-questions
To unsubscribe, send any mail to "freebsd-questions-unsubscr...@freebsd.org"


Re: sysbench / fileio - Linux vs. FreeBSD

2010-06-05 Thread Max Laier
On Saturday 05 June 2010 01:58:35 Adam PAPAI wrote:
> Why FreeBSD is supreme with 1 and 2 thread. And why is it 2 and 3 times
> slower with 4-8-16-32 threads compared to Debian? The first two tests (1
> thread and 2 thread) showed me that FreeBSD is supreme in I/O, but later
> tests showed me, that it can produce horrible I/O.
> 
> How can I tune my disk to make it faster? Is it possible? What is the
> reason of the really slow I/O with more than 4 threads? What do you
> recommend me to do? Why is it damn slow with 8K blocksize?

You may find this interesting:

http://info.iet.unipi.it/~luigi/geom_sched/

Regards,
  Max
___
freebsd-questions@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-questions
To unsubscribe, send any mail to "freebsd-questions-unsubscr...@freebsd.org"


Re: sysbench / fileio - Linux vs. FreeBSD

2010-06-05 Thread Igor Mozolevsky
 /usr/src : zfs with compression enabled
 /usr/src : 386.3MB/s

>>> Do I understand it well? It seems that zfs with compression enabled on
>>> /usr/src with 8KB block size and 16 threads performs 386.3MB/s which
>>> is about 6 times better than debian5? I am thinking about this image
>>> http://tech-blog.wooh.hu/~wooh/debian_vs_freebsd_io_16_seqwr.png
>>
>> Yes - on one run it even hit 500MB/s. I suspect, however, that the
>> benchmark isn't accurate because it won't be writing typical data.
>> Instead it's probably using a buffer that compresses very well.
>
> Hm.. My ZFS tests showed me the same results. With compression it's
> pretty fast.

That's hardly a surprise - you take the source code, compress it into
virtual non-existence leaving hardly anything to be written to the
disk... Obviously if compression speed >> IO speed and the result of
the compression is a significant reduction in size, you have a massive
gain in writing that data to the disk.


--
Igor
___
freebsd-questions@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-questions
To unsubscribe, send any mail to "freebsd-questions-unsubscr...@freebsd.org"


Re: sysbench / fileio - Linux vs. FreeBSD

2010-06-05 Thread Adam PAPAI
-BEGIN PGP SIGNED MESSAGE-
Hash: SHA1

On 6/5/10 1:04 PM, Bruce Cran wrote:
> On Sat, 5 Jun 2010 12:50:15 +0200
> Stefan Miklosovic  wrote:
> 
>>> /var : ufs with softupdates
>>> /usr/obj : zfs with checksums disabled
>>> /usr/src : zfs with compression enabled
>>> /home   : zfs with compression disabled and checksums enabled
>>>
>>> I ran a test with a blocksize of 8KB and 16 threads.
>>>
>>> /var   : 25.2MB/s
>>> /usr/obj : 64.8MB/s
>>> /usr/src : 386.3MB/s
>>> /home   : 60.3MB/s
>>
>> Do I understand it well? It seems that zfs with compression enabled on
>> /usr/src with 8KB block size and 16 threads performs 386.3MB/s which
>> is about 6 times better than debian5? I am thinking about this image
>> http://tech-blog.wooh.hu/~wooh/debian_vs_freebsd_io_16_seqwr.png
> 
> Yes - on one run it even hit 500MB/s. I suspect, however, that the
> benchmark isn't accurate because it won't be writing typical data.
> Instead it's probably using a buffer that compresses very well.

Hm.. My ZFS tests showed me the same results. With compression it's
pretty fast. An application benchmark will give us typical data write,
so I'll run PgSQL benchmarks on the ZFS pool as well.


- -- 
Adam PAPAI
NETIDEA Informatikai Szolgáltató Kft.
http://www.netidea.hu
E-mail: w...@wooh.hu
Phone: +36 30 33-55-735 (Hungary)
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.10 (Darwin)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/

iQEcBAEBAgAGBQJMCjJiAAoJEGq0EWvh5uiIKFMH/1dP4OZGAMiBNSoRqGFfnZ5B
/vtf5do2t3JRbjfYi2HyNn8gXss4xRDPouVmftl2OglIXA77hMIyIcjyoWnHGTBc
M1WnnNDz1wIb8EYSl9MYKAjQA1wGsYd4UImd1MqOtZfSuOht6hTLoSiAnC1xMLtk
9vgFUtMok8XclPqL08J/dWs39+HwhSaooRnLEx7IYLSgFis7vQtJjOaWWG3LUADw
QsivcCSjBBoQ7LD9WXN5prmlwt+CMBU/F1yyMaJXa0bNI7AM+hh5Mix03P4HAKEz
4Z92lcmLXzSVnllA0tAJvAwEPtk4laP6yzM9egStDNvxONLueQVLXfY8gvukQ2k=
=MVI2
-END PGP SIGNATURE-
___
freebsd-questions@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-questions
To unsubscribe, send any mail to "freebsd-questions-unsubscr...@freebsd.org"


Re: sysbench / fileio - Linux vs. FreeBSD

2010-06-05 Thread Bruce Cran
On Sat, 5 Jun 2010 12:50:15 +0200
Stefan Miklosovic  wrote:

> > /var : ufs with softupdates
> > /usr/obj : zfs with checksums disabled
> > /usr/src : zfs with compression enabled
> > /home   : zfs with compression disabled and checksums enabled
> >
> > I ran a test with a blocksize of 8KB and 16 threads.
> >
> > /var   : 25.2MB/s
> > /usr/obj : 64.8MB/s
> > /usr/src : 386.3MB/s
> > /home   : 60.3MB/s
> 
> Do I understand it well? It seems that zfs with compression enabled on
> /usr/src with 8KB block size and 16 threads performs 386.3MB/s which
> is about 6 times better than debian5? I am thinking about this image
> http://tech-blog.wooh.hu/~wooh/debian_vs_freebsd_io_16_seqwr.png

Yes - on one run it even hit 500MB/s. I suspect, however, that the
benchmark isn't accurate because it won't be writing typical data.
Instead it's probably using a buffer that compresses very well.

-- 
Bruce Cran
___
freebsd-questions@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-questions
To unsubscribe, send any mail to "freebsd-questions-unsubscr...@freebsd.org"


Re: sysbench / fileio - Linux vs. FreeBSD

2010-06-05 Thread Stefan Miklosovic
> /var : ufs with softupdates
> /usr/obj : zfs with checksums disabled
> /usr/src : zfs with compression enabled
> /home   : zfs with compression disabled and checksums enabled
>
> I ran a test with a blocksize of 8KB and 16 threads.
>
> /var   : 25.2MB/s
> /usr/obj : 64.8MB/s
> /usr/src : 386.3MB/s
> /home   : 60.3MB/s

Do I understand it well? It seems that zfs with compression enabled on
/usr/src with 8KB block size and 16 threads performs 386.3MB/s which
is about 6 times better than debian5? I am thinking about this image
http://tech-blog.wooh.hu/~wooh/debian_vs_freebsd_io_16_seqwr.png

what is your system specs?
___
freebsd-questions@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-questions
To unsubscribe, send any mail to "freebsd-questions-unsubscr...@freebsd.org"


Re: sysbench / fileio - Linux vs. FreeBSD

2010-06-05 Thread Adam PAPAI
On 6/5/10 3:36 AM, Bruce Cran wrote:
> Some quick tests show that ufs does do rather poorly on my system too. I have 
> the following filesystems setup:
> 
> /var : ufs with softupdates
> /usr/obj : zfs with checksums disabled
> /usr/src : zfs with compression enabled
> /home   : zfs with compression disabled and checksums enabled
> 
> I ran a test with a blocksize of 8KB and 16 threads.
> 
> /var   : 25.2MB/s
> /usr/obj : 64.8MB/s
> /usr/src : 386.3MB/s
> /home   : 60.3MB/s
> 

It seems I have to test it with zfs as well. Tomorrow I'm gonna test it.


-- 
Adam PAPAI
___
freebsd-questions@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-questions
To unsubscribe, send any mail to "freebsd-questions-unsubscr...@freebsd.org"


Re: sysbench / fileio - Linux vs. FreeBSD

2010-06-04 Thread Bruce Cran
On Saturday 05 June 2010 00:58:35 Adam PAPAI wrote:

> Why FreeBSD is supreme with 1 and 2 thread. And why is it 2 and 3 times
> slower with 4-8-16-32 threads compared to Debian? The first two tests (1
> thread and 2 thread) showed me that FreeBSD is supreme in I/O, but later
> tests showed me, that it can produce horrible I/O.
> 
> How can I tune my disk to make it faster? Is it possible? What is the
> reason of the really slow I/O with more than 4 threads? What do you
> recommend me to do? Why is it damn slow with 8K blocksize?

Some quick tests show that ufs does do rather poorly on my system too. I have 
the following filesystems setup:

/var : ufs with softupdates
/usr/obj : zfs with checksums disabled
/usr/src : zfs with compression enabled
/home   : zfs with compression disabled and checksums enabled

I ran a test with a blocksize of 8KB and 16 threads.

/var   : 25.2MB/s
/usr/obj : 64.8MB/s
/usr/src : 386.3MB/s
/home   : 60.3MB/s

-- 
Bruce Cran
___
freebsd-questions@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-questions
To unsubscribe, send any mail to "freebsd-questions-unsubscr...@freebsd.org"


Re: sysbench / fileio - Linux vs. FreeBSD

2010-06-04 Thread Igor Mozolevsky
On 5 June 2010 00:58, Adam PAPAI  wrote:

> How can I tune my disk to make it faster? Is it possible? What is the
> reason of the really slow I/O with more than 4 threads? What do you
> recommend me to do? Why is it damn slow with 8K blocksize?

Does linux still have async disk writes by default?


Igor
___
freebsd-questions@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-questions
To unsubscribe, send any mail to "freebsd-questions-unsubscr...@freebsd.org"


sysbench / fileio - Linux vs. FreeBSD

2010-06-04 Thread Adam PAPAI
-BEGIN PGP SIGNED MESSAGE-
Hash: SHA1

Hi List,

A week ago I started to benchmark Linux vs. FreeBSD on a Dell Poweredge
1850.

CPU: 2 x 3.4Ghz Xeon (Dual Core)
Memory: 8GB (4x2)
Disk: 1 x SEAGATE ST373454LC D404 (SCSI)

FreeBSD kazoku 8.0-RELEASE-p3 FreeBSD 8.0-RELEASE-p3 #0: Tue May 25
20:54:11 UTC 2010
r...@amd64-builder.daemonology.net:/usr/obj/usr/src/sys/GENERIC  amd64

The tests with seqrewr, seqrd, rndrd, and so on is still going on, so I
can only publish the seqwr result. (The PostgreSQL will be tested as well)

(soft-updates are on)
/dev/da0s1d on /usr (ufs, local, soft-updates)

Tested with:
sysbench --num-threads=$a --file-block-size=$bs --test=fileio
- --file-total-size=2G --file-fsync-all=no --file-test-mode=seqwr run

My first results (seqwr with 1,2,4,8,6,32 threads) can be found here.

http://tech-blog.wooh.hu/~wooh/fbsd_vs_debian_seqwr.html

Why FreeBSD is supreme with 1 and 2 thread. And why is it 2 and 3 times
slower with 4-8-16-32 threads compared to Debian? The first two tests (1
thread and 2 thread) showed me that FreeBSD is supreme in I/O, but later
tests showed me, that it can produce horrible I/O.

How can I tune my disk to make it faster? Is it possible? What is the
reason of the really slow I/O with more than 4 threads? What do you
recommend me to do? Why is it damn slow with 8K blocksize?

I have more than 15 FreeBSD servers in production environment and I
don't want to change operating system due to I/O issues. I changed my
OpenBSD servers to FreeBSD 3 years ago... :)

When all tests are ready I'll publish all the results, including the
postgresql benchmarks as well.

Best Regards,

- -- 
Adam PAPAI
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.10 (Darwin)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/

iQEcBAEBAgAGBQJMCZMrAAoJEGq0EWvh5uiI10MIAM1iZxFZ5xssKmawHl56Ruin
zHHgb4Nc15waTLdzFGfllAayDlZqvvpoSpOVbp8qDZYlkTbYPF6aMjkehqMvQUEo
nFs7WN2VaCSOhUUQSwjqfGdnMLW9H5uyW/ZkYvgoOjQjz/vewDV6Fi+ZfGmt5Zqw
gV1ZlXFdAUOUW6c90ODOPxn+7XCA5UC2sUMPB+1iNxrTiiS6C2YQ0Vy1fCXvrhU3
51n0ES/7JBF4sk5dH1VNEU/8AeQRBOoKPuAHhZKRZZ1x+1dMkDhwdD+KUHGrRGJd
fUAZmMhjE6fRG86FbwK5jrZizHZYpE3PfpZe6tI3SIvw7NbUNrRsCMSiel+0FBg=
=k3Sw
-END PGP SIGNATURE-
___
freebsd-questions@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-questions
To unsubscribe, send any mail to "freebsd-questions-unsubscr...@freebsd.org"