Ask Your Question

Revision history [back]

In production, you are likely to want / need different kind of hardware for the different nodes: - For controller : IO + CPU + Network - For network : Network + some CPU - For compute : CPU

Besides that you may also want some kind of redundancy for the critical nodes / services, and you may want to have a lot of compute nodes or scaling the quantity of those.

You'll have different quantities of administration work to do on the different kind of nodes, backup for example, should not even be needed for the compute nodes, because they should already be easily / automatically (re-)constructed, etc...

All of those may be reasons to do it the way it is documented.

It all depends on what exactly you mean by production. If you want to have a few virtualized internal services, or run a full datacenter of computationally intensive science research projects, provide on-demand WEB servers, etc...

In production, you are likely to want / need different kind of hardware for the different nodes: - nodes:

  • For controller : IO + CPU + Network - Network
  • For network : Network + some CPU - CPU
  • For compute : CPU

Besides that you may also want some kind of redundancy for the critical nodes / services, and you may want to have a lot of compute nodes or scaling the quantity of those.

You'll have different quantities of administration work to do on the different kind of nodes, backup for example, should not even be needed for the compute nodes, because they should already be easily / automatically (re-)constructed, etc...

All of those may be reasons to do it the way it is documented.

It all depends on what exactly you mean by production. If you want to have a few virtualized internal services, or run a full datacenter of computationally intensive science research projects, provide on-demand WEB servers, etc...

In production, you are likely to want / need different kind of hardware for the different nodes:

  • For controller : IO + CPU + Network
  • For network : Network + some CPU
  • For compute : CPUCPU (eventually some GPUs too)

Besides that you may also want some kind of redundancy for the critical nodes / services, and you may want to have a lot of compute nodes or scaling the quantity of those.those. The CPUs on your compute nodes are the really interesting ones, you may want to have some ARM or PowerPC VMs running. It's there that the CPU's hardware virtualization features will be needed, so you'll want the youngest generation for those nodes.

You'll have different quantities of administration work to do on the different kind of nodes, backup for example, should not even be needed for the compute nodes, because they should already be easily / automatically (re-)constructed, (re-)constructed through automation (puppet, chef, salt...), etc...

All of those may be reasons to do it the way it is documented.

It all depends on what exactly you mean by production. If you want to have a few virtualized internal services, or run a full datacenter of computationally intensive science research projects, provide on-demand WEB servers, etc...

In production, you are likely to want / need different kind of hardware for the different nodes:

  • For controller : IO + CPU + Network
  • For network : Network + some CPU
  • For compute : CPU + RAM (eventually some GPUs too)

Besides that you may also want some kind of redundancy for the critical nodes / services, and you may want to have a lot of compute nodes or scaling the quantity of those. The CPUs on your compute nodes are the really interesting ones, you may want to have some ARM or PowerPC VMs running. It's there that the CPU's hardware virtualization features will be needed, so you'll want the youngest generation for those nodes.

You'll have different quantities of administration work to do on the different kind of nodes, backup for example, should not even be needed for the compute nodes, because they should already be easily / automatically (re-)constructed through automation (puppet, chef, salt...), etc...

All of those may be reasons to do it the way it is documented.

It all depends on what exactly you mean by production. If you want to have a few virtualized internal services, or run a full datacenter of computationally intensive science research projects, provide on-demand WEB servers, etc...