You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@cloudstack.apache.org by Yiping Zhang <yz...@marketo.com> on 2016/03/08 23:10:34 UTC

XenServer cluster size

Hi, all:

The CloudStack doc recommends that for XenServer, do not put more than 8 hosts in a cluster, while the Citrix XenServer doc says that XenServer 6.5 can natively support 16 hosts in a cluster (resource pool).

I am wondering why CloudStack is recommending a smaller cluster size than that XenServer can natively support?  If I create a cluster with 10 XenServers, what could go wrong for me ?  Has any one tried with CS cluster with >8 XenServer hosts ?

My environment is CS 4.5.1 (soon to be upgraded to 4.8.0) on RHEL 6.7 and XenServer 6.5, using NetApp volumes for both primary and secondary storages.

Yiping

Re: XenServer cluster size

Posted by Yiping Zhang <yz...@marketo.com>.
I think I didn’t make it clear.  Those 10 hosts are all Gen8 blades, that’s why I want them to form a pool of their own.  New Gen9 blades will go into a different pool.



On 3/9/16, 10:14 AM, "Tim Mackey" <tm...@gmail.com> wrote:

>In that case, Yiping, I would *definitely* recommend putting those servers
>into at least two pools. The processors used in Gen8 and Gen9 servers can
>not currently be joined into the same pool, and you actually need to be
>very sensitive to the processor steppings. Dundee should fix that, but no
>current version of CloudStack supports Dundee (and neither does Citrix at
>the moment).
>
>-tim
>
>On Wed, Mar 9, 2016 at 12:38 PM, Yiping Zhang <yz...@marketo.com> wrote:
>
>> Hi, Tim:
>>
>> Thanks for very detailed reply.
>>
>> These are Gen8 HP blades and all my new servers will be Gen9.  That’s why
>> I’d like to combine them into one maxed out cluster.  I have only two guest
>> VLAN’s and roughly 400 VM instances for this 10 hosts cluster.  So I think
>> performance wise I should be OK.
>>
>> Yiping
>>
>>
>>
>> On 3/8/16, 4:28 PM, "Tim Mackey" <tm...@gmail.com> wrote:
>>
>> >Yiping,
>> >
>> >Here's the detailed answer ....
>> >
>> >From the XenServer perspective, there are a number of factors which go
>> into
>> >how various configuration limits are arrived at. Most of the time, they
>> >aren't hard limits (for example I know of users with more than 16 hosts in
>> >a pool). What the XenServer team do is for a given metric they determine
>> >the point at which overall scalability is reduced to a target threshold.
>> >That then becomes the "configuration limit" for a given release, and we
>> >retest with every version.
>> >
>> >In the case of the "hosts per pool" limit, we need to ensure that all
>> >operations we have can be performed without impairment with a given number
>> >of hosts in a pool. We've kept the same maximum number of hosts in a pool
>> >for a very long time (close to ten years so far), and that's a direct
>> >reflection of how much we've increased individual host scalability.
>> >
>> >From a CloudStack perspective, there have been a number of serious scale
>> >limits which have pushed XenServer. Hundreds of VLANs is one example that
>> >Ahmad cites, but its also a case of the number of VMs and needing to
>> manage
>> >all those VM objects.  iirc, the eight host recommendation came from some
>> >large deployment requirements. If you don't have a need for 100s of VLANs
>> >per pool, or aren't running 100s of VMs per host, you likely will be able
>> >to get more than eight hosts per pool.
>> >
>> >From an operations perspective, I would look closely at your pool size and
>> >ask the question of why you want to such a large pool.  I'd argue having
>> >two pools of five hosts is more efficient in CloudStack than a single pool
>> >of ten hosts, plus if something should happen to one pool, the remaining
>> >pool will continue to be available.  CloudStack is very efficient at
>> >managing resource pools, so many of the reasons traditional server admins
>> >cite for wanting large pool sizes aren't as relevant in CloudStack.
>> >
>> >Of particular note is how you scale. With a ten host pool size, that's
>> your
>> >scalability block size, so as you grow you'll want to increase capacity in
>> >chunks of ten hosts. With a smaller pool size, you'd be able to add
>> >capacity in much smaller chunks.
>> >
>> >-tim
>> >
>> >On Tue, Mar 8, 2016 at 5:24 PM, Ahmad Emneina <ae...@gmail.com> wrote:
>> >
>> >> IIRC, its just a recommendation. I think it stemmed from performance
>> >> impact, due to numerous VLAN's present, in environments with lots of
>> >> tenants.
>> >>
>> >> On Tue, Mar 8, 2016 at 2:10 PM, Yiping Zhang <yz...@marketo.com>
>> wrote:
>> >>
>> >> > Hi, all:
>> >> >
>> >> > The CloudStack doc recommends that for XenServer, do not put more
>> than 8
>> >> > hosts in a cluster, while the Citrix XenServer doc says that XenServer
>> >> 6.5
>> >> > can natively support 16 hosts in a cluster (resource pool).
>> >> >
>> >> > I am wondering why CloudStack is recommending a smaller cluster size
>> than
>> >> > that XenServer can natively support?  If I create a cluster with 10
>> >> > XenServers, what could go wrong for me ?  Has any one tried with CS
>> >> cluster
>> >> > with >8 XenServer hosts ?
>> >> >
>> >> > My environment is CS 4.5.1 (soon to be upgraded to 4.8.0) on RHEL 6.7
>> and
>> >> > XenServer 6.5, using NetApp volumes for both primary and secondary
>> >> storages.
>> >> >
>> >> > Yiping
>> >> >
>> >>
>>

Re: XenServer cluster size

Posted by Tim Mackey <tm...@gmail.com>.
In that case, Yiping, I would *definitely* recommend putting those servers
into at least two pools. The processors used in Gen8 and Gen9 servers can
not currently be joined into the same pool, and you actually need to be
very sensitive to the processor steppings. Dundee should fix that, but no
current version of CloudStack supports Dundee (and neither does Citrix at
the moment).

-tim

On Wed, Mar 9, 2016 at 12:38 PM, Yiping Zhang <yz...@marketo.com> wrote:

> Hi, Tim:
>
> Thanks for very detailed reply.
>
> These are Gen8 HP blades and all my new servers will be Gen9.  That’s why
> I’d like to combine them into one maxed out cluster.  I have only two guest
> VLAN’s and roughly 400 VM instances for this 10 hosts cluster.  So I think
> performance wise I should be OK.
>
> Yiping
>
>
>
> On 3/8/16, 4:28 PM, "Tim Mackey" <tm...@gmail.com> wrote:
>
> >Yiping,
> >
> >Here's the detailed answer ....
> >
> >From the XenServer perspective, there are a number of factors which go
> into
> >how various configuration limits are arrived at. Most of the time, they
> >aren't hard limits (for example I know of users with more than 16 hosts in
> >a pool). What the XenServer team do is for a given metric they determine
> >the point at which overall scalability is reduced to a target threshold.
> >That then becomes the "configuration limit" for a given release, and we
> >retest with every version.
> >
> >In the case of the "hosts per pool" limit, we need to ensure that all
> >operations we have can be performed without impairment with a given number
> >of hosts in a pool. We've kept the same maximum number of hosts in a pool
> >for a very long time (close to ten years so far), and that's a direct
> >reflection of how much we've increased individual host scalability.
> >
> >From a CloudStack perspective, there have been a number of serious scale
> >limits which have pushed XenServer. Hundreds of VLANs is one example that
> >Ahmad cites, but its also a case of the number of VMs and needing to
> manage
> >all those VM objects.  iirc, the eight host recommendation came from some
> >large deployment requirements. If you don't have a need for 100s of VLANs
> >per pool, or aren't running 100s of VMs per host, you likely will be able
> >to get more than eight hosts per pool.
> >
> >From an operations perspective, I would look closely at your pool size and
> >ask the question of why you want to such a large pool.  I'd argue having
> >two pools of five hosts is more efficient in CloudStack than a single pool
> >of ten hosts, plus if something should happen to one pool, the remaining
> >pool will continue to be available.  CloudStack is very efficient at
> >managing resource pools, so many of the reasons traditional server admins
> >cite for wanting large pool sizes aren't as relevant in CloudStack.
> >
> >Of particular note is how you scale. With a ten host pool size, that's
> your
> >scalability block size, so as you grow you'll want to increase capacity in
> >chunks of ten hosts. With a smaller pool size, you'd be able to add
> >capacity in much smaller chunks.
> >
> >-tim
> >
> >On Tue, Mar 8, 2016 at 5:24 PM, Ahmad Emneina <ae...@gmail.com> wrote:
> >
> >> IIRC, its just a recommendation. I think it stemmed from performance
> >> impact, due to numerous VLAN's present, in environments with lots of
> >> tenants.
> >>
> >> On Tue, Mar 8, 2016 at 2:10 PM, Yiping Zhang <yz...@marketo.com>
> wrote:
> >>
> >> > Hi, all:
> >> >
> >> > The CloudStack doc recommends that for XenServer, do not put more
> than 8
> >> > hosts in a cluster, while the Citrix XenServer doc says that XenServer
> >> 6.5
> >> > can natively support 16 hosts in a cluster (resource pool).
> >> >
> >> > I am wondering why CloudStack is recommending a smaller cluster size
> than
> >> > that XenServer can natively support?  If I create a cluster with 10
> >> > XenServers, what could go wrong for me ?  Has any one tried with CS
> >> cluster
> >> > with >8 XenServer hosts ?
> >> >
> >> > My environment is CS 4.5.1 (soon to be upgraded to 4.8.0) on RHEL 6.7
> and
> >> > XenServer 6.5, using NetApp volumes for both primary and secondary
> >> storages.
> >> >
> >> > Yiping
> >> >
> >>
>

Re: XenServer cluster size

Posted by Yiping Zhang <yz...@marketo.com>.
Hi, Tim:

Thanks for very detailed reply.

These are Gen8 HP blades and all my new servers will be Gen9.  That’s why I’d like to combine them into one maxed out cluster.  I have only two guest VLAN’s and roughly 400 VM instances for this 10 hosts cluster.  So I think performance wise I should be OK.

Yiping



On 3/8/16, 4:28 PM, "Tim Mackey" <tm...@gmail.com> wrote:

>Yiping,
>
>Here's the detailed answer ....
>
>From the XenServer perspective, there are a number of factors which go into
>how various configuration limits are arrived at. Most of the time, they
>aren't hard limits (for example I know of users with more than 16 hosts in
>a pool). What the XenServer team do is for a given metric they determine
>the point at which overall scalability is reduced to a target threshold.
>That then becomes the "configuration limit" for a given release, and we
>retest with every version.
>
>In the case of the "hosts per pool" limit, we need to ensure that all
>operations we have can be performed without impairment with a given number
>of hosts in a pool. We've kept the same maximum number of hosts in a pool
>for a very long time (close to ten years so far), and that's a direct
>reflection of how much we've increased individual host scalability.
>
>From a CloudStack perspective, there have been a number of serious scale
>limits which have pushed XenServer. Hundreds of VLANs is one example that
>Ahmad cites, but its also a case of the number of VMs and needing to manage
>all those VM objects.  iirc, the eight host recommendation came from some
>large deployment requirements. If you don't have a need for 100s of VLANs
>per pool, or aren't running 100s of VMs per host, you likely will be able
>to get more than eight hosts per pool.
>
>From an operations perspective, I would look closely at your pool size and
>ask the question of why you want to such a large pool.  I'd argue having
>two pools of five hosts is more efficient in CloudStack than a single pool
>of ten hosts, plus if something should happen to one pool, the remaining
>pool will continue to be available.  CloudStack is very efficient at
>managing resource pools, so many of the reasons traditional server admins
>cite for wanting large pool sizes aren't as relevant in CloudStack.
>
>Of particular note is how you scale. With a ten host pool size, that's your
>scalability block size, so as you grow you'll want to increase capacity in
>chunks of ten hosts. With a smaller pool size, you'd be able to add
>capacity in much smaller chunks.
>
>-tim
>
>On Tue, Mar 8, 2016 at 5:24 PM, Ahmad Emneina <ae...@gmail.com> wrote:
>
>> IIRC, its just a recommendation. I think it stemmed from performance
>> impact, due to numerous VLAN's present, in environments with lots of
>> tenants.
>>
>> On Tue, Mar 8, 2016 at 2:10 PM, Yiping Zhang <yz...@marketo.com> wrote:
>>
>> > Hi, all:
>> >
>> > The CloudStack doc recommends that for XenServer, do not put more than 8
>> > hosts in a cluster, while the Citrix XenServer doc says that XenServer
>> 6.5
>> > can natively support 16 hosts in a cluster (resource pool).
>> >
>> > I am wondering why CloudStack is recommending a smaller cluster size than
>> > that XenServer can natively support?  If I create a cluster with 10
>> > XenServers, what could go wrong for me ?  Has any one tried with CS
>> cluster
>> > with >8 XenServer hosts ?
>> >
>> > My environment is CS 4.5.1 (soon to be upgraded to 4.8.0) on RHEL 6.7 and
>> > XenServer 6.5, using NetApp volumes for both primary and secondary
>> storages.
>> >
>> > Yiping
>> >
>>

Re: XenServer cluster size

Posted by Tim Mackey <tm...@gmail.com>.
Yiping,

Here's the detailed answer ....

>From the XenServer perspective, there are a number of factors which go into
how various configuration limits are arrived at. Most of the time, they
aren't hard limits (for example I know of users with more than 16 hosts in
a pool). What the XenServer team do is for a given metric they determine
the point at which overall scalability is reduced to a target threshold.
That then becomes the "configuration limit" for a given release, and we
retest with every version.

In the case of the "hosts per pool" limit, we need to ensure that all
operations we have can be performed without impairment with a given number
of hosts in a pool. We've kept the same maximum number of hosts in a pool
for a very long time (close to ten years so far), and that's a direct
reflection of how much we've increased individual host scalability.

>From a CloudStack perspective, there have been a number of serious scale
limits which have pushed XenServer. Hundreds of VLANs is one example that
Ahmad cites, but its also a case of the number of VMs and needing to manage
all those VM objects.  iirc, the eight host recommendation came from some
large deployment requirements. If you don't have a need for 100s of VLANs
per pool, or aren't running 100s of VMs per host, you likely will be able
to get more than eight hosts per pool.

>From an operations perspective, I would look closely at your pool size and
ask the question of why you want to such a large pool.  I'd argue having
two pools of five hosts is more efficient in CloudStack than a single pool
of ten hosts, plus if something should happen to one pool, the remaining
pool will continue to be available.  CloudStack is very efficient at
managing resource pools, so many of the reasons traditional server admins
cite for wanting large pool sizes aren't as relevant in CloudStack.

Of particular note is how you scale. With a ten host pool size, that's your
scalability block size, so as you grow you'll want to increase capacity in
chunks of ten hosts. With a smaller pool size, you'd be able to add
capacity in much smaller chunks.

-tim

On Tue, Mar 8, 2016 at 5:24 PM, Ahmad Emneina <ae...@gmail.com> wrote:

> IIRC, its just a recommendation. I think it stemmed from performance
> impact, due to numerous VLAN's present, in environments with lots of
> tenants.
>
> On Tue, Mar 8, 2016 at 2:10 PM, Yiping Zhang <yz...@marketo.com> wrote:
>
> > Hi, all:
> >
> > The CloudStack doc recommends that for XenServer, do not put more than 8
> > hosts in a cluster, while the Citrix XenServer doc says that XenServer
> 6.5
> > can natively support 16 hosts in a cluster (resource pool).
> >
> > I am wondering why CloudStack is recommending a smaller cluster size than
> > that XenServer can natively support?  If I create a cluster with 10
> > XenServers, what could go wrong for me ?  Has any one tried with CS
> cluster
> > with >8 XenServer hosts ?
> >
> > My environment is CS 4.5.1 (soon to be upgraded to 4.8.0) on RHEL 6.7 and
> > XenServer 6.5, using NetApp volumes for both primary and secondary
> storages.
> >
> > Yiping
> >
>

Re: XenServer cluster size

Posted by Ahmad Emneina <ae...@gmail.com>.
IIRC, its just a recommendation. I think it stemmed from performance
impact, due to numerous VLAN's present, in environments with lots of
tenants.

On Tue, Mar 8, 2016 at 2:10 PM, Yiping Zhang <yz...@marketo.com> wrote:

> Hi, all:
>
> The CloudStack doc recommends that for XenServer, do not put more than 8
> hosts in a cluster, while the Citrix XenServer doc says that XenServer 6.5
> can natively support 16 hosts in a cluster (resource pool).
>
> I am wondering why CloudStack is recommending a smaller cluster size than
> that XenServer can natively support?  If I create a cluster with 10
> XenServers, what could go wrong for me ?  Has any one tried with CS cluster
> with >8 XenServer hosts ?
>
> My environment is CS 4.5.1 (soon to be upgraded to 4.8.0) on RHEL 6.7 and
> XenServer 6.5, using NetApp volumes for both primary and secondary storages.
>
> Yiping
>