The Systems Alchemist
Serving People with Technology-- Not the other way around.
Friday, March 26, 2010
VCP4 - Passed
Thanks to some incredible training last month, I was able to pass my VCP4 exam this week. After working with VMware ESX (etc.) for the last 2-3 years, it's nice to finally have my knowledge and experience certified.
Thursday, March 18, 2010
SMP Results
The theory looks like it was correct...I dropped a VM from 2 CPUs to 1 CPU and saw a 10x reduction in CPU Ready times. The server went from 2500 ms wait times to 250 ms almost instantaneously.
I'm still waiting to hear from my client to make sure that this is translating into better performance, as he perceives it, but this is definitely a step in the right direction as far as helping our client use their existing resources more efficiently.
I'm still waiting to hear from my client to make sure that this is translating into better performance, as he perceives it, but this is definitely a step in the right direction as far as helping our client use their existing resources more efficiently.
Update from client: "The overall performance is much better thank you. I do still think that there is some room for improvement, but I’m no longer swearing at the machine which is a great thing."
Tuesday, March 9, 2010
Performance: SMP on ESX
I'm working with a large vSphere4 environment currently that has multiple vCPUs assigned to almost every VM. This includes web servers and application development/testing servers...
After learning more about SMP, it makes sense that "CPU Ready" latency is high on almost all of the VMs...likely the result of "Idle" cycles being queued up on the physical processors for VMs that are only utilizing single CPU threads.
I'm working with an application owner to drop the number of CPUs to 1 in the hope that it will help me make a case for reducing the CPU count across the board, except for in cases that are explicitly proven to require more than one.
More updates to follow after the adjustment...
After learning more about SMP, it makes sense that "CPU Ready" latency is high on almost all of the VMs...likely the result of "Idle" cycles being queued up on the physical processors for VMs that are only utilizing single CPU threads.
I'm working with an application owner to drop the number of CPUs to 1 in the hope that it will help me make a case for reducing the CPU count across the board, except for in cases that are explicitly proven to require more than one.
More updates to follow after the adjustment...
Labels:
esx,
optimization,
performance,
vmware,
vsphere
Subscribe to:
Posts (Atom)
