Dryv
Enthusiast
Enthusiast

vCPU/pCPU


Hi Guys,

I have a very basic question and I'm a litlte confused at the abstraction layer I guess. If I have server with 2 Physical CPUs with 10 cores each hyperthreaded, this gives me 40 vCPUs to play with... I think?

Now, just for understanding purposes... if I wanted to install a single VM only on this server... what do I set its cpu settings to so it can (if it wants) consume as much as the hardware can offer?

I'm really struggling on understanding the basics here.

Thanks

Dryv

7 Replies
rcporto
Leadership
Leadership

I have a very basic question and I'm a litlte confused at the abstraction layer I guess. If I have server with 2 Physical CPUs with 10 cores each hyperthreaded, this gives me 40 vCPUs to play with... I think?

Yes, you're right about the numbers of logical processors that your ESXi will see.

About the single VM, this blog post may answer your question: When to Overcommit vCPU:pCPU for Monster VMs | VMware vSphere Blog - VMware Blogs

---

Richardson Porto
Senior Infrastructure Specialist
LinkedIn: http://linkedin.com/in/richardsonporto
Alistar
Expert
Expert

If you want to fully utilize the ESXi host, set your Physical CPU : virtual CPU ratio to 1:1 without hyperthreading. As earlier stated, HT is there to help "offload" some threads within one single execution tick. To comply with the NUMA architecture, I'd say you go with 2physical sockets and 10 cores per socket on the VM to comply with the NUMA architecture the host is running on top of.

Stop by my blog if you'd like 🙂 I dabble in vSphere troubleshooting, PowerCLI scripting and NetApp storage - and I share my journeys at http://vmxp.wordpress.com/
0 Kudos
vThinkBeyondVM
VMware Employee
VMware Employee

Please read "ESXi CPU considerations section (page 20) from below white paper:

http://www.vmware.com/pdf/Perf_Best_Practices_vSphere5.5.pdf

If it is useful, plz mark answer as correct or helpful. ---------------------------------------------------------------- Thanks & Regards Vikas, VCP50, MCTS on AD, SCJP6.0. http://vThinkBeyondVM.com ----------------------------------------------------------------- Disclaimer: Any views or opinions expressed here are strictly my own. I am solely responsible for all content published here. Content published here is not read, reviewed or approved in advance by VMware and does not necessarily represent or reflect the views or opinions of VMware.
0 Kudos
Dryv
Enthusiast
Enthusiast

Hi Guys

Thanks for the great responses and additional reading material. Just to take my analogy a little further, please keep me honest. Let's say I wanted to grant my VM the whole resources of the underlying host and protect it using FT. Now as we know FT currently has a constraint of one vcpu... Could I:

1. In a hyper threaded environment grant the VM within its settings 1 vcpu and 40 cores?

2. In a non hyper-threaded environment grant the VM within its settings 1 vcpu and 20 cores?

And meet the FT constraint...? Is this something that would be considered a real newbie mistake to make... ? Please educate me.... The experience of the board is much appreciated.

Dryv

0 Kudos
rcporto
Leadership
Leadership

No matter the way you create your virtual machine, if the total number of core is greater than 1, you cannot use FT to protect this virtual machine.

---

Richardson Porto
Senior Infrastructure Specialist
LinkedIn: http://linkedin.com/in/richardsonporto
a_p_
Leadership
Leadership

Unfortunately not. FT currently only supports a single vCPU/vCore.

André

Dryv
Enthusiast
Enthusiast

Okay... That helps me in understanding that! Thanks chaps.

So... Let's try and crack this one with something like the Stratus product... Has anyone got any experiences with it?

As you have probably determined I need to protect a vendor supplied application cross site... 100% uptime in a DR scenario... It has been coded in isolation over the years with no thought around making it distributed across multiple sites... Causes us endless amount of pain and grief... In a DR scenario I need to keep it running as it holds a lot in memory so a reboot would be no good... Otherwise I could simply protect this using SRM with an underlying synch replicated lun..... Happy days.

0 Kudos