Hi,
i have a windows 10 VM running on a M820 blade server it has 12vCPU & 128GB ram. the os is installed on a Samsung SSD evo. but im abit stuck as autocad is really really slow, when you connect to data it takes ages and ages for the grid coordinates to load up but before when ive done it without a GPU it works immediately. is it possible something to do with too many CPU or RAM, or something else.
Matt
When the previous time it worked immediately was the configuration the same?
Anyway, possible explanations:
(1) too many vCPUs in the Windows 10 (although this is unlikely to be the cause)
You can check whether all vCPU within the Windows 10 VM are busy while loading. If only 1 or 2 vCPUs are busy while the 10-11 vCPUs are idle/near idle (close to 0% utilization), there is no advantage to having the additional vCPUs. There is a perverse effect of having too many vCPUs where most are idle in the VM. The idle instructions of the vCPU will still have to be executed in the host CPU and therefore take away the execution time slots from the actual vCPU that need to have its instructions executed on the host CPU. It is a bit like having 10-11 extra people blocking/disturbing 1-2 people who are doing actual work. Although this sort of perverse situation should have been addressed by a CPU feature called Paused-Loop Exiting that came with Westmere generation of CPUs and ESXi 6.5 should be able to take advantage of this feature.
(2) Meltdown patch to the Windows 10 VM
If you already have the Meltdown patch applied to the Windows 10 VM, it will likely experience some slow down especially with I/O intensive tasks (disk I/O, network I/O). Given that the CPU you have is 2.7GHz E5-4650 (Sandy Bridge), it does NOT have the INVPCID (INValidate PCID) instruction. The INVPCID is available on Haswell generation and later CPUs. Windows 10 will not be able to make use of PCID (Process Context Identifier) in the Translation Lookaside Buffer (TLB) if the CPU/vCPU does not have INVPCID to mitigate performance hits with the Meltdown patch. Without PCID/INVPCID, the performance hit come as the TLB needs to be flushed constantly with every context switch and I/O intensive tasks tend to have many process context switches. If you run the Get-SpeculationControlSettings Powershell within the VM it will show the PCID performance mitigation as FALSE.
On ESXi host run ESXTOP and check CPU Ready% and Co-Stop values. These values should be low; high values indicate CPU overprovisioning (the VM is waiting for available CPU cores)
can you please share more information about ESXi host version and build number and guest VM configuration .
Hi ranchuba,
added a screenshot of the Host Config's and VM information.
Matt
When the previous time it worked immediately was the configuration the same?
Anyway, possible explanations:
(1) too many vCPUs in the Windows 10 (although this is unlikely to be the cause)
You can check whether all vCPU within the Windows 10 VM are busy while loading. If only 1 or 2 vCPUs are busy while the 10-11 vCPUs are idle/near idle (close to 0% utilization), there is no advantage to having the additional vCPUs. There is a perverse effect of having too many vCPUs where most are idle in the VM. The idle instructions of the vCPU will still have to be executed in the host CPU and therefore take away the execution time slots from the actual vCPU that need to have its instructions executed on the host CPU. It is a bit like having 10-11 extra people blocking/disturbing 1-2 people who are doing actual work. Although this sort of perverse situation should have been addressed by a CPU feature called Paused-Loop Exiting that came with Westmere generation of CPUs and ESXi 6.5 should be able to take advantage of this feature.
(2) Meltdown patch to the Windows 10 VM
If you already have the Meltdown patch applied to the Windows 10 VM, it will likely experience some slow down especially with I/O intensive tasks (disk I/O, network I/O). Given that the CPU you have is 2.7GHz E5-4650 (Sandy Bridge), it does NOT have the INVPCID (INValidate PCID) instruction. The INVPCID is available on Haswell generation and later CPUs. Windows 10 will not be able to make use of PCID (Process Context Identifier) in the Translation Lookaside Buffer (TLB) if the CPU/vCPU does not have INVPCID to mitigate performance hits with the Meltdown patch. Without PCID/INVPCID, the performance hit come as the TLB needs to be flushed constantly with every context switch and I/O intensive tasks tend to have many process context switches. If you run the Get-SpeculationControlSettings Powershell within the VM it will show the PCID performance mitigation as FALSE.
Can I ask how you are using that Win10 vm? Horizon Desktop to publish ?
hi dave,
I am viewing it via RDP on a thin client. But it also is just the same on the VMware remote console off vcenter.
matt