Skip to content

Instantly share code, notes, and snippets.

Created June 23, 2016 13:37
Show Gist options
  • Save anonymous/fe1c153dec0b37794f0c807bdb8ca9dc to your computer and use it in GitHub Desktop.
Save anonymous/fe1c153dec0b37794f0c807bdb8ca9dc to your computer and use it in GitHub Desktop.
VMware Technology Day 2016 Taiwan
  • CPU Performance
    • Ready time need to know
      • High CPU usage -> Best Virtualization utilization for enterprise, not really the CPU resource is exhausted
    • Higher Ready time -> Wait for Physical CPU Ready for a long time -> may be the CPU exhausted with High CPU usage
    • resxtop
      • V for show vm only
      • e for show the world of specified vm
    • Not to use SMP (1 CPU 1 vCPU)
  • Memory Performance
    • vm memory usage strategy: on demend -> will not return back memory
      • Transparent page sharing has security issue, so default disabled, but can be enabled if needed
      • Balloon
      • Memory compressed
      • Host-level swapping -> will degrade the vm performance extremely -> need to check
    • consumption memory vs. Active memory
    • resxtop/esxtop
      • m to show memory usage
      • state
        • high or medium use ballon or not
        • Low use host-level swapping or vm doesn't have balloon driver (vmtools)
  • Network performance
    • emulation and sharing
      • vmtools need to be installed, virtual network adapter will be shared, not emulated
      • vmxnet3 / vmxnet2 / vmxnet (determined by host nic)
    • Jumbo frames
    • resxtop
      • n to show network stats
    • make baseline
      • make baseline when somedays which nothing happened, average or statistics
    • drop network packets
    • Drop received packets
      • guest os cpu performance low
    • drop transmit packets
  • Storage performance
    • Determined the requests
    • queue and lun queue depth
      • max recommended queue depth is 64
    • Disk throughput
      • resxtop
        • d to show disk stat
        • u to show queue stat
        • Actv and qued
    • Command abords for overloaded storage, must be 0
    • Slow storage
      • DAVG on average > 10 ms and peak > 20 ms
  • vSAN
    • in one disk group, vmware prefer ssd:hd = 1:10
    • Availability
      • vm mirroring: number of replication
      • Striping: slice one vm into multiple part on disks
      • All can mix to perform
    • Flash read cache reservation
      • on ssd read cache:write buffer=7:3, fixed
      • For critical mission, reserv read cache to boost performance, and let other mission degrade performance
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment