Re: Cores from Multiple Physical Hosts in VM

2013-11-16 Thread Rafael Weingartner
It is a hypervisor restriction not CS.

I can wonder some work around like using a distributed shared memory system
and on top of it installing a hypervisor like Xen and XCP. Then, it would
work since the DSM would simulate a single machine with the resource of all
machines on the cluster.

Of course, there is a trade of, DSM generates a pretty good overhead, but it
is interesting to try it out and check the performance.


2013/11/16 Robert Gabriel epheme...@gmail.com

 On 15 November 2013 15:47, m...@kelceydamage.com m...@kelceydamage.com
 wrote:

  The only people I have seen that have macro VMs working was some company
  out of Boston making virtual router/networking services with special
  interconnects that allowed process/thread striping over multiple
  hosts/sockets.
 
  Sent from my HTC
 
  - Reply message -
  From: Sebastien Goasguen run...@gmail.com
  To: users@cloudstack.apache.org
  Subject: Cores from Multiple Physical Hosts in VM
  Date: Fri, Nov 15, 2013 2:08 AM
 
  On Nov 13, 2013, at 7:03 AM, Robert Gabriel epheme...@gmail.com wrote:
 
   Hi,
  
   Is it possible to do the below?
  
   Thank you.
  
  
   Answered by sgordon:
  
   It is my understanding that this is not currently possible, there was
  some
   discussion at the design summit (I think in the Libvirt driver roadmap
   session) about making the scheduler NUMA aware which would allow such
   configurations on hardware that supports NUMA but this is currently
   unimplemented.
  
   In reply to ephemeric's question: Cores from Multiple Physical Hosts in
  VM
  
   Tags: vcpus, aggregates, hosts, multiple, physical.
  
   Hi,
  
   Pardon my ignorance as I have never looked at cloud computing.
  
   Is it possible to create a VM and assign to it cores from multiple
  physical
   hosts for high vcpu numbers?
  
   We have the following problem: Splunk running 38 concurrent searches
 on a
   blade that only has 16 cores.
  
   By creating a VM and combining the cores from two blades, hence 32
 vcpus
  in
   total somehow?
 
  As far as I know this is not possible.
 
  That said I would be very surprise if Splunk could not use multiple
  machines. So just run multiple instances (separate VMs) that point to the
  same data store.
 
  
   I'm not sure if this is possible.
  
   Thank you.
  
   To change frequency, language and content of these alerts, please visit
  your
   user profile
  https://ask.openstack.org/en/users/2044/ephemeric/subscriptions/.
  
  
   If you believe that this message was sent in an error, please email
 about
   it the forum administrator at communitym...@openstack.org.
 

 Thank you, at least gives me a direction.




-- 
Rafael Weingärtner


Re: Cores from Multiple Physical Hosts in VM

2013-11-15 Thread m...@kelceydamage.com
The only people I have seen that have macro VMs working was some company out of 
Boston making virtual router/networking services with special interconnects 
that allowed process/thread striping over multiple hosts/sockets.

Sent from my HTC

- Reply message -
From: Sebastien Goasguen run...@gmail.com
To: users@cloudstack.apache.org
Subject: Cores from Multiple Physical Hosts in VM
Date: Fri, Nov 15, 2013 2:08 AM

On Nov 13, 2013, at 7:03 AM, Robert Gabriel epheme...@gmail.com wrote:

 Hi,
 
 Is it possible to do the below?
 
 Thank you.
 
 
 Answered by sgordon:
 
 It is my understanding that this is not currently possible, there was some
 discussion at the design summit (I think in the Libvirt driver roadmap
 session) about making the scheduler NUMA aware which would allow such
 configurations on hardware that supports NUMA but this is currently
 unimplemented.
 
 In reply to ephemeric's question: Cores from Multiple Physical Hosts in VM
 
 Tags: vcpus, aggregates, hosts, multiple, physical.
 
 Hi,
 
 Pardon my ignorance as I have never looked at cloud computing.
 
 Is it possible to create a VM and assign to it cores from multiple physical
 hosts for high vcpu numbers?
 
 We have the following problem: Splunk running 38 concurrent searches on a
 blade that only has 16 cores.
 
 By creating a VM and combining the cores from two blades, hence 32 vcpus in
 total somehow?

As far as I know this is not possible. 

That said I would be very surprise if Splunk could not use multiple machines. 
So just run multiple instances (separate VMs) that point to the same data store.

 
 I'm not sure if this is possible.
 
 Thank you.
 
 To change frequency, language and content of these alerts, please visit your
 user 
 profilehttps://ask.openstack.org/en/users/2044/ephemeric/subscriptions/.
 
 
 If you believe that this message was sent in an error, please email about
 it the forum administrator at communitym...@openstack.org.

Re: Cores from Multiple Physical Hosts in VM

2013-11-15 Thread Robert Gabriel
On 15 November 2013 15:47, m...@kelceydamage.com m...@kelceydamage.com wrote:

 The only people I have seen that have macro VMs working was some company
 out of Boston making virtual router/networking services with special
 interconnects that allowed process/thread striping over multiple
 hosts/sockets.

 Sent from my HTC

 - Reply message -
 From: Sebastien Goasguen run...@gmail.com
 To: users@cloudstack.apache.org
 Subject: Cores from Multiple Physical Hosts in VM
 Date: Fri, Nov 15, 2013 2:08 AM

 On Nov 13, 2013, at 7:03 AM, Robert Gabriel epheme...@gmail.com wrote:

  Hi,
 
  Is it possible to do the below?
 
  Thank you.
 
 
  Answered by sgordon:
 
  It is my understanding that this is not currently possible, there was
 some
  discussion at the design summit (I think in the Libvirt driver roadmap
  session) about making the scheduler NUMA aware which would allow such
  configurations on hardware that supports NUMA but this is currently
  unimplemented.
 
  In reply to ephemeric's question: Cores from Multiple Physical Hosts in
 VM
 
  Tags: vcpus, aggregates, hosts, multiple, physical.
 
  Hi,
 
  Pardon my ignorance as I have never looked at cloud computing.
 
  Is it possible to create a VM and assign to it cores from multiple
 physical
  hosts for high vcpu numbers?
 
  We have the following problem: Splunk running 38 concurrent searches on a
  blade that only has 16 cores.
 
  By creating a VM and combining the cores from two blades, hence 32 vcpus
 in
  total somehow?

 As far as I know this is not possible.

 That said I would be very surprise if Splunk could not use multiple
 machines. So just run multiple instances (separate VMs) that point to the
 same data store.

 
  I'm not sure if this is possible.
 
  Thank you.
 
  To change frequency, language and content of these alerts, please visit
 your
  user profile
 https://ask.openstack.org/en/users/2044/ephemeric/subscriptions/.
 
 
  If you believe that this message was sent in an error, please email about
  it the forum administrator at communitym...@openstack.org.


Thank you, at least gives me a direction.


Cores from Multiple Physical Hosts in VM

2013-11-13 Thread Robert Gabriel
Hi,

Is it possible to do the below?

Thank you.


Answered by sgordon:

It is my understanding that this is not currently possible, there was some
discussion at the design summit (I think in the Libvirt driver roadmap
session) about making the scheduler NUMA aware which would allow such
configurations on hardware that supports NUMA but this is currently
unimplemented.

In reply to ephemeric's question: Cores from Multiple Physical Hosts in VM

Tags: vcpus, aggregates, hosts, multiple, physical.

Hi,

Pardon my ignorance as I have never looked at cloud computing.

Is it possible to create a VM and assign to it cores from multiple physical
hosts for high vcpu numbers?

We have the following problem: Splunk running 38 concurrent searches on a
blade that only has 16 cores.

By creating a VM and combining the cores from two blades, hence 32 vcpus in
total somehow?

I'm not sure if this is possible.

Thank you.

To change frequency, language and content of these alerts, please visit your
user profilehttps://ask.openstack.org/en/users/2044/ephemeric/subscriptions/.


If you believe that this message was sent in an error, please email about
it the forum administrator at communitym...@openstack.org.