0

Virtualization experts! I have a question that revolves around a potential CPU virtualization concept illustrated below. Is there an existing implementation of this idea, or are there hardware limitations? Specifically, I'm curious if any of you have encountered CPU virtualization where the software on Server 1 can amalgamate CPU cores from nodes (node 1 & 2) or slave machines to create a vCPU with a significantly higher core count. The goal is for programs running on Server 1 to perceive a larger number of CPU cores (e.g., 16) rather than the actual 4 or 8.

While I'm aware that Big Data may employ a similar approach, it often involves breaking down programs into a map-reduce mechanism. My query is whether there's a mechanism within the virtualization software itself to automatically break down program executables into chunks, distribute them to node CPUs (with data stored in memory), and execute these chunks. This process would ideally be handled by the virtualization software, eliminating the need for manual programming.

I appreciate your insights on this matter. Feel free to share your comments, and if you know of any other relevant groups or discussion forums, please point me in the right direction.

enter image description here

pmaniyan
  • 101

1 Answers1

2

Yes, exists. performance was abysmal and required high end hardware - talk multiple 100g network links. It is also - especially given your numbers - a joke. 16 cores are a high-end DESKTOP server. My personal servers have 2x16 - and they are ending end of life. Always cheaper to get proper hardware. For anything larger, the network latency, memory access etc. are just going to break you. Comes up every other year and generally shows a lack of fundamental understanding how a computer operates internally by the person asking.

While I'm aware that Big Data may employ a similar approach, it often. involves breaking down programs into a map-reduce mechanism.

No. It ALWAYS involves different computers OPERATING AS SEPARATE COMPUTERS with software that works together.

The ONLY scenario where I see something like this is in GPU AI Training. And the hardware side here is brutal - we talk of servers with terabit network connectivity (20+ 100g links). And guess what - the SOFTWARE stack is optimized for this, which you demand it not to be.

TomTom
  • 52,109
  • 7
  • 59
  • 142