LVM performance problems with large number of VGs

LVM performance problems with large number of VGs

[Syed Mushtaq]

[-- Attachment #1.1: Type: text/plain, Size: 885 bytes --]

Hi,

I am testing out a system in which each Guest disk is mapped to a LUN on a
backend storage. I am using Xenserver 6.5 for this and creating an SR with
a single VDI inside it for each Guest disk. The way it would work is, from
an LVM point of view, I would have a Volume Group (VG) for each virtual
disk and a single logical volume (LV) inside that VG. What I am seeing is
that creating new volume groups is taking a lot of time. In my test after
creating about 600 volume groups, it took about 4 seconds to create a new
one. I have observed this to be a linear trend where the more VGs I add the
slower future operations become. Other operations like getting the logical
volumes inside a VG have also slowed down (even though there is only 1 LV
per VG).

I was wondering if anyone here faced a similar problem or had some
experience in dealing with LVM bottlenecks.

Thanks,
-Syed

[-- Attachment #1.2: Type: text/html, Size: 978 bytes --]

[-- Attachment #2: Type: text/plain, Size: 126 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

Re: LVM performance problems with large number of VGs

[Wei Liu]

On Fri, Apr 01, 2016 at 10:23:42AM -0400, Syed Mushtaq wrote:
> Hi,
> 
> I am testing out a system in which each Guest disk is mapped to a LUN on a
> backend storage. I am using Xenserver 6.5 for this and creating an SR with
> a single VDI inside it for each Guest disk. The way it would work is, from
> an LVM point of view, I would have a Volume Group (VG) for each virtual
> disk and a single logical volume (LV) inside that VG. What I am seeing is
> that creating new volume groups is taking a lot of time. In my test after
> creating about 600 volume groups, it took about 4 seconds to create a new
> one. I have observed this to be a linear trend where the more VGs I add the
> slower future operations become. Other operations like getting the logical
> volumes inside a VG have also slowed down (even though there is only 1 LV
> per VG).
> 
> I was wondering if anyone here faced a similar problem or had some
> experience in dealing with LVM bottlenecks.
> 

Hi this mailing list is for upstream Xen development. For XenServer
related stuff please go to xenserver.org and look for correct channel
there.

Wei.

> Thanks,
> -Syed

> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xen.org
> http://lists.xen.org/xen-devel


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

Re: LVM performance problems with large number of VGs

[Syed Mushtaq]

[-- Attachment #1.1: Type: text/plain, Size: 1739 bytes --]

I understand that Xenserver has a different channel but I thought of
mailing here as it would target a wider audience and I believe the problem
that I am facing would also surface itself within upstream Xen as well as
Xen also uses LVM as one of its storage options and the problem is very
specific to LVM.

Thanks,
-Syed


On Fri, Apr 1, 2016 at 11:22 AM, Wei Liu <wei.liu2@citrix.com> wrote:

> On Fri, Apr 01, 2016 at 10:23:42AM -0400, Syed Mushtaq wrote:
> > Hi,
> >
> > I am testing out a system in which each Guest disk is mapped to a LUN on
> a
> > backend storage. I am using Xenserver 6.5 for this and creating an SR
> with
> > a single VDI inside it for each Guest disk. The way it would work is,
> from
> > an LVM point of view, I would have a Volume Group (VG) for each virtual
> > disk and a single logical volume (LV) inside that VG. What I am seeing is
> > that creating new volume groups is taking a lot of time. In my test after
> > creating about 600 volume groups, it took about 4 seconds to create a new
> > one. I have observed this to be a linear trend where the more VGs I add
> the
> > slower future operations become. Other operations like getting the
> logical
> > volumes inside a VG have also slowed down (even though there is only 1 LV
> > per VG).
> >
> > I was wondering if anyone here faced a similar problem or had some
> > experience in dealing with LVM bottlenecks.
> >
>
> Hi this mailing list is for upstream Xen development. For XenServer
> related stuff please go to xenserver.org and look for correct channel
> there.
>
> Wei.
>
> > Thanks,
> > -Syed
>
> > _______________________________________________
> > Xen-devel mailing list
> > Xen-devel@lists.xen.org
> > http://lists.xen.org/xen-devel
>
>

[-- Attachment #1.2: Type: text/html, Size: 2447 bytes --]

[-- Attachment #2: Type: text/plain, Size: 126 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

Re: LVM performance problems with large number of VGs

[Roger Pau Monné]

On Fri, 1 Apr 2016, Syed Mushtaq wrote:
> Hi,
> 
> I am testing out a system in which each Guest disk is mapped to a LUN on a backend storage. I am using Xenserver 6.5 for this and creating an SR with a single VDI inside it for each Guest disk.
> The way it would work is, from an LVM point of view, I would have a Volume Group (VG) for each virtual disk and a single logical volume (LV) inside that VG. What I am seeing is that creating new
> volume groups is taking a lot of time. In my test after creating about 600 volume groups, it took about 4 seconds to create a new one. I have observed this to be a linear trend where the more
> VGs I add the slower future operations become. Other operations like getting the logical volumes inside a VG have also slowed down (even though there is only 1 LV per VG).
> 
> I was wondering if anyone here faced a similar problem or had some experience in dealing with LVM bottlenecks.

It doesn't seem like this is related to Xen, so I would recommend that you 
ask in the linux-lvm [0] mailing list. TBH, I have no idea how well LVM 
scales in these kind of scenarios.

Roger.

[0] https://www.redhat.com/mailman/listinfo/linux-lvm

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

Re: LVM performance problems with large number of VGs

[Syed Mushtaq]

[-- Attachment #1.1: Type: text/plain, Size: 1372 bytes --]

Thanks Roger. Will do that.

-Syed

On Fri, Apr 1, 2016 at 12:38 PM, Roger Pau Monné <roger.pau@citrix.com>
wrote:

> On Fri, 1 Apr 2016, Syed Mushtaq wrote:
> > Hi,
> >
> > I am testing out a system in which each Guest disk is mapped to a LUN on
> a backend storage. I am using Xenserver 6.5 for this and creating an SR
> with a single VDI inside it for each Guest disk.
> > The way it would work is, from an LVM point of view, I would have a
> Volume Group (VG) for each virtual disk and a single logical volume (LV)
> inside that VG. What I am seeing is that creating new
> > volume groups is taking a lot of time. In my test after creating about
> 600 volume groups, it took about 4 seconds to create a new one. I have
> observed this to be a linear trend where the more
> > VGs I add the slower future operations become. Other operations like
> getting the logical volumes inside a VG have also slowed down (even though
> there is only 1 LV per VG).
> >
> > I was wondering if anyone here faced a similar problem or had some
> experience in dealing with LVM bottlenecks.
>
> It doesn't seem like this is related to Xen, so I would recommend that you
> ask in the linux-lvm [0] mailing list. TBH, I have no idea how well LVM
> scales in these kind of scenarios.
>
> Roger.
>
> [0] https://www.redhat.com/mailman/listinfo/linux-lvm
>

[-- Attachment #1.2: Type: text/html, Size: 1836 bytes --]

[-- Attachment #2: Type: text/plain, Size: 126 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel