Sign in with
Sign up | Sign in
Your question
Solved

Setting up a cluster to work as one in Windows Server 2012 R2

Tags:
  • Clustering
  • Windows Server
  • Computers
  • Windows Server 2012
  • Nodes
  • Business Computing
Last response: in Business Computing
Share
August 15, 2014 11:08:47 AM

Hopefully this is the appropriate thread to ask this question. My question is how can I take four identical computers and use them as one? So essentially it would have 8 total cores if they were dual core processors.

I've already setup Windows Server 2012 with Active Directory and a domain controller on a Dell PowerEdge R310. I've used the HPC software to link the 4 computers in question to the Dell R310 (the Headnode), thats all fine and good but its not accomplishing what I want it to. Based on the searches i've done on google it looks like people generally are using some form of linux to accomplish this. Not that I have anything against Linux I just would like to try and accomplish this on windows.

I would imagine I would have to setup a virtual machine to be distributed across the four computers so I then could install an one OS for the four computers. Also from my search on the internet, I am getting mixed responses, some people say it is possible, others say this isn't possible. Is this possible with Hyper-V that is built in with Windows Server 2012?

I am looking for straight forward answers and an explanation with that answer.

Thank you to whoever takes time to try and resolve my question.

More about : setting cluster work windows server 2012

August 22, 2014 9:18:07 AM

There is LIMITED capability for a VM to span multiple physical hosts http://en.wikipedia.org/wiki/Parallel_Virtual_Machine . But I don't think this is what you are looking for. Clustered computers usually have to have software apps written for a distributed environment. If you were using something like MatLab, you can get it to run with distributed compute nodes. Autocad -- has add on libraries to do distributed 3D rendering. But off the shelf software, not going to happen.
m
0
l
Related resources

Best solution

August 23, 2014 3:33:28 PM

Flappers67 said:
Hopefully this is the appropriate thread to ask this question. My question is how can I take four identical computers and use them as one? So essentially it would have 8 total cores if they were dual core processors.

I've already setup Windows Server 2012 with Active Directory and a domain controller on a Dell PowerEdge R310. I've used the HPC software to link the 4 computers in question to the Dell R310 (the Headnode), thats all fine and good but its not accomplishing what I want it to. Based on the searches i've done on google it looks like people generally are using some form of linux to accomplish this. Not that I have anything against Linux I just would like to try and accomplish this on windows.

I would imagine I would have to setup a virtual machine to be distributed across the four computers so I then could install an one OS for the four computers. Also from my search on the internet, I am getting mixed responses, some people say it is possible, others say this isn't possible. Is this possible with Hyper-V that is built in with Windows Server 2012?

I am looking for straight forward answers and an explanation with that answer.

Thank you to whoever takes time to try and resolve my question.


This thread is a bit old, but what you are looking for is distributed computing configuration known as a Single System Image. In an SSI configuration the resources of all the nodes are pooled, any node can access the resources available on any other node. Logically, all of the nodes in an SSI appear to form one big computer and this same illusion is exposed to running processes. This hides the complexity of the cluster configuration but processes that are unaware of the fact that the underlying hardware is distributed may cause problems.
SSI configurations are simpler to administer than other cluster configurations but have some severe drawbacks. The biggest one is that the interconnect method can introduce a massive bottleneck. RDMA is often used to access memory across nodes, so high performance RDMA technology such as Infiniband work well at masking interconnect bottlenecks but more common interconnects such as RDMA over Ethernet do not. As such, great care needs to be taken at multiple levels to avoid crippling application performance.
As a result of the complexity involved in squeezing performance out of SSI most vendors have shifted to using a head node to schedule and distribute tasks to a number of dedicated compute nodes and load balancers. This avoids the problems associated with cross-node resource access and simplifies task design but increases the complexity of administration.

If you simply want to muck around with SSI, give OpenSSI a try.
Share
August 25, 2014 3:59:11 PM

Pinhedd said:
Flappers67 said:
Hopefully this is the appropriate thread to ask this question. My question is how can I take four identical computers and use them as one? So essentially it would have 8 total cores if they were dual core processors.

I've already setup Windows Server 2012 with Active Directory and a domain controller on a Dell PowerEdge R310. I've used the HPC software to link the 4 computers in question to the Dell R310 (the Headnode), thats all fine and good but its not accomplishing what I want it to. Based on the searches i've done on google it looks like people generally are using some form of linux to accomplish this. Not that I have anything against Linux I just would like to try and accomplish this on windows.

I would imagine I would have to setup a virtual machine to be distributed across the four computers so I then could install an one OS for the four computers. Also from my search on the internet, I am getting mixed responses, some people say it is possible, others say this isn't possible. Is this possible with Hyper-V that is built in with Windows Server 2012?

I am looking for straight forward answers and an explanation with that answer.

Thank you to whoever takes time to try and resolve my question.


This thread is a bit old, but what you are looking for is distributed computing configuration known as a Single System Image. In an SSI configuration the resources of all the nodes are pooled, any node can access the resources available on any other node. Logically, all of the nodes in an SSI appear to form one big computer and this same illusion is exposed to running processes. This hides the complexity of the cluster configuration but processes that are unaware of the fact that the underlying hardware is distributed may cause problems.
SSI configurations are simpler to administer than other cluster configurations but have some severe drawbacks. The biggest one is that the interconnect method can introduce a massive bottleneck. RDMA is often used to access memory across nodes, so high performance RDMA technology such as Infiniband work well at masking interconnect bottlenecks but more common interconnects such as RDMA over Ethernet do not. As such, great care needs to be taken at multiple levels to avoid crippling application performance.
As a result of the complexity involved in squeezing performance out of SSI most vendors have shifted to using a head node to schedule and distribute tasks to a number of dedicated compute nodes and load balancers. This avoids the problems associated with cross-node resource access and simplifies task design but increases the complexity of administration.

If you simply want to muck around with SSI, give OpenSSI a try.


Thanks so much man! I'll definitely give that a try!
m
0
l
!