Re: [linux-lvm] Performance tunning on LVM2

[Heinz Mauelshagen <mauelshagen@redhat.com>]

On Mon, Jun 09, 2008 at 12:06:19PM +0200, Antony MARTINEAU wrote:
> Thanks for your answer...
> But my test show that it is, the LVM2 software the probleme...

device-mapper snapshot target actually.

> 
> Because even whith 3 writing test at the same time on 3 LV which are on 
> the same VG and the same PV, the performance are better than one LV whith 
> only one snapshot...

Sure, the write patterns for snapshots go sequentially to the disk
(i.e. read from origin, write to COW store, write to origin, ...).

> 
> Look this test:
> 
> suse2:~ # dd if=/dev/zero of=/dev/vg0/test bs=10M count=150
> 150+0 records in
> 150+0 records out
> 1572864000 bytes (1.6 GB) copied, 27.9422 seconds, 56.3 MB/s
> 
> suse2:~ # dd if=/dev/zero of=/dev/vg0/test2 bs=10M count=150
> 150+0 records in
> 150+0 records out
> 1572864000 bytes (1.6 GB) copied, 33.2836 seconds, 47.3 MB/s
> 
> suse2:~ # dd if=/dev/zero of=/dev/vg0/test3 bs=10M count=150
> 150+0 records in
> 150+0 records out
> 1572864000 bytes (1.6 GB) copied, 33.784 seconds, 46.6 MB/s
> 
> With 3 writing test AT THE SAME TIME , the average is better that one 
> writing test on one LV whith one snap
> 
> Look,
> 
> suse2:~ # lvcreate -s -L2G -ntest.snap /dev/vg0/test
>   Logical volume "test.snap" created
> 
> suse2:~ # dd if=/dev/zero of=/dev/vg0/test bs=10M count=150
> 150+0 records in
> 150+0 records out
> 1572864000 bytes (1.6 GB) copied, 382.315 seconds, 4.1 MB/s
> 
> It is disastrous...
> 
> I think dd is a good test...

It's an extreme test like I tried to point out.

Like already mentioned: you gotta put the COW store on a seperate PV
after adding one to your vg0 (say /dev/sdb1).

E.g.:
pvcreate /dev/sdb1
vgextend vg0 /dev/sdb1
lvcreate -s -L2G -ntest.snap /dev/vg0/test /dev/sdb1

Heinz


> 
> 
> Cordialement,
> 
> 
> 
> MARTINEAU
> Antony 
> Service informatique 
> Assistant informatique 
> LIPPI Management 
> La Fouillouse
> 16440 Mouthiers sur Boheme 
> Tel.: 05.45.67.34.35
> Courriel: antony.martineau@lippi.fr
> http://www.lippi.fr 
> 
> 
> 
> 
> De :
> Heinz Mauelshagen <mauelshagen@redhat.com>
> Pour :
> LVM general discussion and development <linux-lvm@redhat.com>
> Date:
> 09/06/2008 11:46
> Objet :
> Re: [linux-lvm] Performance tunning on LVM2
> 
> 
> 
> On Fri, Jun 06, 2008 at 08:03:51AM -0700, Larry Dickson wrote:
> > A (linear) volume group made of two physical volumes consists of one PV
> > followed by the other, rather like a "Raid-Linear". If you size the
> > origin logical volume right, you can get one LV (the origin) to fall on 
> one
> > disk, and force the snapshot to land on the other disk. This eliminates
> > back-and-forth seeking to the COW. Whether it solves your problem will
> > depend on how smart the driver is about the read-before-write activity 
> on
> > the origin volume.
> > 
> > Other members of the list may have more experience on this. Comments?
> 
> If I read correctly, Antony just has *ONE* PV.
> 
> So no matter what, he has to add another to allow for snapshot COW
> store allocation on that other PV, distinct from the one holding
> the origin(s). Presumably there's no other bottleneck aside from the
> disk, that'll do better.
> 
> Keep in mind, that unless you've got streaming writes, the performance
> won't drop as much as in the (artificial) dd test below.
> 
> FYI: With the current snapshot implementation, multiple snapshots per 
> single
>      origin will throttle write performance because of write duplication
>      to all per snapshot COW stores.
> 
> Heinz
> 
> > 
> > Larry
> > 
> > On 6/6/08, Antony MARTINEAU <Antony.MARTINEAU@lippi.fr> wrote:
> > >
> > >
> > > The volume group vg0 is the raid0 of two disk (SAS 15000rpm 300G0)
> > > I have only this raid on the server
> > >
> > > But i don't understand, imagine i make a volume group  ou of this 
> raid0. It
> > > is no possible to snapshot the original volume, am i wrong?
> > >
> > > If i make a new VG on another disks, For exemple /dev/vg1/
> > > LVM don't permit to store a snaphot on a different VG than the origin
> > > volum.
> > >
> > > for exemple /dev/vg0/test cant be snapshoting on /dev/vg1/test.snap
> > >
> > > LV test and LV test.snap must be on the same volume, am i wrong ???? 
> so it
> > > is impossible to store snapshot on another disk....
> > >
> > >
> > >   Cordialement,
> > >
> > >    *MARTINEAU
> > > Antony*
> > > Service informatique
> > > Assistant informatique
> > > LIPPI Management La Fouillouse
> > > 16440 Mouthiers sur Boheme
> > > Tel.: 05.45.67.34.35
> > > Courriel: *antony.martineau@lippi.fr* <antony.martineau@lippi.fr>*
> > > **http://www.lippi.fr* <http://www.lippi.fr/>
> > >
> > >
> > >
> > >   De : "Larry Dickson" <ldickson@cuttedge.com> Pour : "LVM general
> > > discussion and development" <linux-lvm@redhat.com> Date: 06/06/2008 
> 16:19 Objet
> > > : Re: [linux-lvm] Performance tunning on LVM2
> > > ------------------------------
> > >
> > >
> > >
> > > This looks like the result of excessive seeking. Are origin volume and
> > > snapshot both on the same physical drive? Is it possible to make a 
> volume
> > > group out of two drives, and arrange things so that origin volume and
> > > snapshot are hitting different disks?
> > >
> > > Larry Dickson
> > > Cutting Edge Networked Storage
> > >
> > > On 6/6/08, *Antony MARTINEAU* 
> <*Antony.MARTINEAU@lippi.fr*<Antony.MARTINEAU@lippi.fr>>
> > > wrote:
> > >
> > > Hello,
> > > My configuration:
> > > Server DELL 2860 Intel(R) Xeon(R) CPU  X3230 @ 2.66GHz (Quad Core)
> > > 8GB of  memory
> > > 2 x SAS 15000 300G0 RAID 0 hardware
> > > SLES 10 SP2
> > > Kernel 2.6.16.60-0.21-xen
> > >
> > > i have one volume group vg0 ( whith one PV, the two disks in raid0) 
> whith
> > > many lvm
> > > I am very surprise about LVM2 performance when a snapshot is done.
> > > Write speed on the Original volume is very bad when a snaphot is 
> active...
> > >
> > > For exemple:
> > > *
> > > Speed on /dev/vg0/test when there is NO snapshot :*
> > >
> > > suse2:~ # dd if=/dev/zero of=/dev/vg0/test bs=2M count=400
> > > 400+0 records in
> > > 400+0 records out
> > > 838860800 bytes (839 MB) copied, 6.42741 seconds, 131 MB/s
> > > *
> > > Speed on /dev/vg0/test when there is one snapshot of this original 
> volume :
> > > *
> > >
> > > suse2:~ # lvremove --force /dev/vg0/test3.snap
> > >  Logical volume "test3.snap" successfully removed
> > > suse2:~ # dd if=/dev/zero of=/dev/vg0/test bs=2M count=400
> > > 400+0 records in
> > > 400+0 records out
> > > 838860800 bytes (839 MB) copied, 6.42741 seconds, 131 MB/s
> > > suse2:~ # lvcreate -s -L1G -ntest.snap /dev/vg0/test
> > >  Logical volume "test.snap" created
> > > suse2:~ # dd if=/dev/zero of=/dev/vg0/test bs=2M count=400
> > > 400+0 records in
> > > 400+0 records out
> > > 838860800 bytes (839 MB) copied, 204.862 seconds, 4.1 MB/s
> > >
> > > *
> > > Speed on /dev/vg0/test when there is 2 snapshots of this original 
> volume :
> > > *
> > >
> > > suse2:~ # lvcreate -s -L1G -ntest1.snap /dev/vg0/test
> > >  Logical volume "test1.snap" created
> > > suse2:~ # lvcreate -s -L1G -ntest2.snap /dev/vg0/test
> > >  Logical volume "test2.snap" created
> > > suse2:~ # lvremove /dev/vg0/test2.snap
> > > Do you really want to remove active logical volume "test2.snap"? 
> [y/n]: y
> > >  Logical volume "test2.snap" successfully removed
> > > suse2:~ # dd if=/dev/zero of=/dev/vg0/test bs=2M count=400
> > > 400+0 records in
> > > 400+0 records out
> > > 838860800 bytes (839 MB) copied, 270.928 seconds, 3.1 MB/s
> > >
> > >
> > > Do you know  some elements about tunning performance?,?
> > >
> > > Performances are disastrous when a snaphot is active
> > > Could you give your speed result? and your amelioration??
> > >
> > > ps:Results are the same whithout Kernel Xen and whith a kernel more 
> recent
> > > (*2.6.24.2* <http://2.6.24.2/>)  Cordialement,
> > >    *MARTINEAU
> > > Antony*
> > > Service informatique
> > > Assistant informatique
> > > LIPPI Management La Fouillouse
> > > 16440 Mouthiers sur Boheme
> > > Tel.: 05.45.67.34.35
> > > Courriel: *antony.martineau@lippi.fr* <antony.martineau@lippi.fr>*
> > > **http://www.lippi.fr* <http://www.lippi.fr/>
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > > Ce message et toutes les pieces jointes sont etablis a l'attention
> > > exclusive de ses destinataires et sont strictement confidentiels. 
> *Pour en
> > > savoir plus cliquer ici* <http://www.lippi.fr/disclaimer.php>
> > >
> > > This message and any attachments are confidential to the ordinary user 
> of
> > > the e-mail address to which it was addressed and may also be 
> privileged. *More
> > > information* <http://www.lippi.fr/disclaimer.php>
> > >
> > >
> > > _______________________________________________
> > > linux-lvm mailing list*
> > > **linux-lvm@redhat.com* <linux-lvm@redhat.com>*
> > > **https://www.redhat.com/mailman/listinfo/linux-lvm*<
> https://www.redhat.com/mailman/listinfo/linux-lvm>
> > > read the LVM HOW-TO at *http://tldp.org/HOWTO/LVM-HOWTO/*<
> http://tldp.org/HOWTO/LVM-HOWTO/>
> > > _______________________________________________
> > > linux-lvm mailing list
> > > linux-lvm@redhat.com
> > > https://www.redhat.com/mailman/listinfo/linux-lvm
> > > read the LVM HOW-TO at http://tldp.org/HOWTO/LVM-HOWTO/
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > > Ce message et toutes les pieces jointes sont etablis a l'attention
> > > exclusive de ses destinataires et sont strictement confidentiels. 
> *Pour en
> > > savoir plus cliquer ici* <http://www.lippi.fr/disclaimer.php>
> > >
> > > This message and any attachments are confidential to the ordinary user 
> of
> > > the e-mail address to which it was addressed and may also be 
> privileged. *More
> > > information* <http://www.lippi.fr/disclaimer.php>
> > >
> > >
> > > _______________________________________________
> > > linux-lvm mailing list
> > > linux-lvm@redhat.com
> > > https://www.redhat.com/mailman/listinfo/linux-lvm
> > > read the LVM HOW-TO at http://tldp.org/HOWTO/LVM-HOWTO/
> > >
> 
> 
> 
> 
> > _______________________________________________
> > linux-lvm mailing list
> > linux-lvm@redhat.com
> > https://www.redhat.com/mailman/listinfo/linux-lvm
> > read the LVM HOW-TO at http://tldp.org/HOWTO/LVM-HOWTO/
> 
> =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
> 
> Heinz Mauelshagen                                 Red Hat GmbH
> Consulting Development Engineer                   Am Sonnenhang 11
> Storage Development                               56242 Marienrachdorf
>                                                   Germany
> Mauelshagen@RedHat.com                            PHONE +49  171 7803392
>                                                   FAX   +49 2626 924446
> =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
> 
> _______________________________________________
> linux-lvm mailing list
> linux-lvm@redhat.com
> https://www.redhat.com/mailman/listinfo/linux-lvm
> read the LVM HOW-TO at http://tldp.org/HOWTO/LVM-HOWTO/
> 
> 
> 
> 
> 
> 
> 
> 
> 
> 
> Ce message et toutes les pieces jointes sont etablis a l'attention 
> exclusive de ses destinataires et sont strictement confidentiels. Pour en 
> savoir plus cliquer ici
> 
> This message and any attachments are confidential to the ordinary user of 
> the e-mail address to which it was addressed and may also be privileged. 
> More information

=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-

Heinz Mauelshagen                                 Red Hat GmbH
Consulting Development Engineer                   Am Sonnenhang 11
Storage Development                               56242 Marienrachdorf
                                                  Germany
Mauelshagen@RedHat.com                            PHONE +49  171 7803392
                                                  FAX   +49 2626 924446
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-