You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@cloudstack.apache.org by Sanjeev Neelarapu <sa...@citrix.com> on 2013/07/11 11:44:24 UTC

FW: Regarding the vm.allocation.algorithm.


From: Kiran Koneti
Sent: Thursday, July 11, 2013 3:13 PM
To: #Cloud - Engineering
Subject: Regarding the vm.allocation.algorithm.

Hi All,

Where can we find the documentation for the "vm.allocation.algorithm" global setting parameter.

I just wanted to know the behavior of the different values for this parameter. I see the below values but want to know what is the functionality of each.

1)random
2)firstfit
3)userdispersing
4)userconcentratedpod_firstfit
5)userconcentratedpod_random


Regards,
Kiran.

RE: FW: Regarding the vm.allocation.algorithm.

Posted by Sanjeev Neelarapu <sa...@citrix.com>.
Thanks Prasanna. I was trying to understand the impact of these global setting parameters on the dedicated resources feature.

-----Original Message-----
From: Prasanna Santhanam [mailto:tsp@apache.org] 
Sent: Thursday, July 11, 2013 4:40 PM
To: dev@cloudstack.apache.org
Subject: Re: FW: Regarding the vm.allocation.algorithm.

This is how the planners determine where to place the VM based on needs like enough capacity(cpu, memory, storage)

Once the available capacity and requested capacity is determined, the following heuristics are applied to determine the final deployment plan of the VM:

1. random will choose a random host for the VM (note: random doesn't imply it won't go to the same host. that may appear as an aberration)

2. firstfit as the name suggests will put the VM in the first suitable host that is available with enough capacity

3. userdispersing - choose the cluster such that other VMs of the user
(account) are placed as far apart as possible. As far apart being different clusters - different hosts

4. and 5. are related - concentration basically is the inverse of dispersion. try and put the VM as packed and close as possible wrt other VMs of the user(account). So VM1 and VM2 are said to be concentrated if they belong to the same pod.

The need for 3 arose from providing basic tolerance to failures in physical resources. And the need for 4. and 5. arose so as to provide better performance (read, write, network) to VMs in a group.

Apart from all this there are affinity/anti-affinity groups. Host and Storage tags, HA dedicated hosts, HA tags, etc that can determine the placement/planning of VMs.

Under what context are you looking to understand this though? That can help explain the algorithms involved further.

--
Prasanna.,

On Thu, Jul 11, 2013 at 09:44:24AM +0000, Sanjeev Neelarapu wrote:
> 
> 
> From: Kiran Koneti
> Sent: Thursday, July 11, 2013 3:13 PM
> To: #Cloud - Engineering
> Subject: Regarding the vm.allocation.algorithm.
> 
> Hi All,
> 
> Where can we find the documentation for the "vm.allocation.algorithm" global setting parameter.
> 
> I just wanted to know the behavior of the different values for this parameter. I see the below values but want to know what is the functionality of each.
> 
> 1)random
> 2)firstfit
> 3)userdispersing
> 4)userconcentratedpod_firstfit
> 5)userconcentratedpod_random
> 
> 
> Regards,
> Kiran.


------------------------
Powered by BigRock.com


Re: FW: Regarding the vm.allocation.algorithm.

Posted by Prasanna Santhanam <ts...@apache.org>.
This is how the planners determine where to place the VM based on
needs like enough capacity(cpu, memory, storage)

Once the available capacity and requested capacity is determined, the
following heuristics are applied to determine the final deployment
plan of the VM:

1. random will choose a random host for the VM (note: random doesn't
imply it won't go to the same host. that may appear as an aberration)

2. firstfit as the name suggests will put the VM in the first suitable
host that is available with enough capacity

3. userdispersing - choose the cluster such that other VMs of the user
(account) are placed as far apart as possible. As far apart being
different clusters - different hosts

4. and 5. are related - concentration basically is the inverse of
dispersion. try and put the VM as packed and close as possible wrt
other VMs of the user(account). So VM1 and VM2 are said to be
concentrated if they belong to the same pod.

The need for 3 arose from providing basic tolerance to failures in
physical resources. And the need for 4. and 5. arose so as to provide
better performance (read, write, network) to VMs in a group.

Apart from all this there are affinity/anti-affinity groups. Host and
Storage tags, HA dedicated hosts, HA tags, etc that can determine the
placement/planning of VMs.

Under what context are you looking to understand this though? That can
help explain the algorithms involved further.

-- 
Prasanna.,

On Thu, Jul 11, 2013 at 09:44:24AM +0000, Sanjeev Neelarapu wrote:
> 
> 
> From: Kiran Koneti
> Sent: Thursday, July 11, 2013 3:13 PM
> To: #Cloud - Engineering
> Subject: Regarding the vm.allocation.algorithm.
> 
> Hi All,
> 
> Where can we find the documentation for the "vm.allocation.algorithm" global setting parameter.
> 
> I just wanted to know the behavior of the different values for this parameter. I see the below values but want to know what is the functionality of each.
> 
> 1)random
> 2)firstfit
> 3)userdispersing
> 4)userconcentratedpod_firstfit
> 5)userconcentratedpod_random
> 
> 
> Regards,
> Kiran.


------------------------
Powered by BigRock.com