View Only
  • 1.  Proper VM sizing post P2V

    Posted Aug 23, 2021 06:18 PM

    Physical server shows Peak CPU % is 83% out of a net clock speed of 20.4Ghz.     The CPU of the blade server we are using as P2V destination is 3.1GHz.   Peak memory usage is 8062008KB (around 8GB)    

    How many vCPUs would you give to this VM?   and why?  

    How much memory would you allocate to the VM?  

  • 2.  RE: Proper VM sizing post P2V

    Posted Aug 24, 2021 08:21 AM

    Without any details about the OS, and the applications to know whether they benefit more from a higher CPU clock, or more cores it's more of guessing what may be a proper sizing.

    Anyway, I'd probably start with 8 vCPUs (e.g. two 4-core vCPUs, depending on the guest OS's capabilities), which should be fine from the clock speed, as well as from the number of cores. Regarding memory, I'd likely start with 16GB unless you are low on host/physical memory.


    PS: Discussion moved to Converter Standalone Discussions

  • 3.  RE: Proper VM sizing post P2V

    Posted Aug 24, 2021 01:06 PM

    The stats come from live optics, the tool Dell and other VM partners use for sizing out solutions.  83% of 20.4 is just under 17GHz, divided by 3.1GHz I see 5.4 CPUs as needed to handle the peak load, but certainly would want some more head room.  This is my math, but I was really looking for what others might do in this case in regards to sizing methods.   

    I like your end result, but in a shared infrastructure the blades are what they are and there isn't going to be any separate cluster for CPUs with higher clock speeds, btw I think 3.1 is very fast, much faster than the typical clusters I've seen in the past 15 years of supporting virtualization.    

    Also, not certain about your CPU recommendation given that there is a best practice for allocating CPU which would indicate 1 socket and 8 cores.  

    For the memory allocation you've almost doubled what's used at peak.  I like that you are conservative in your approach but I also wonder if you are overly conservative.  We have all seen the wasted resource reports from the various tools like vROPs and foglight etc etc which indicate you've allocated too much.. that nobody ever seems to pay that much attention to.   

    the intention of my post is to generate discussion around sizing, how folks approach it and such, to see if my methods are aligned with that of others