Improve Utilization on ESX Hosts
Posted by craig
- on May 17th, 2009 in Data Center, Tips, Virtualization | 2 Comments »

As virtualization does help us to consolidate virtual machines into physical servers, we are always looking into a better consolidation ratio and hardware utilization. We may not want to achieve this by hitting to performance issue on the virtual infrastructure. Here is 1 of the way you may consider to improve the utilization on your ESX host.
As my environment, the virtual infrastructure are supporting our global business in Asia Pacific, EMEA and NA time zone. When I tracked back the performance history and found the busy hour for most of the machines will be different. Due to the time zone different, I notice that the hardware utilization on the ESX host are always remain 40 to 50 % all the time.
With the data record on hand, I decided to re-allocate the virtual machines in my virtual machine with VMotion to improve utilization on our virtual infrastructure. I had balanced allocated the Asia Pacific machine mix with Europe and North America servers on every esx host. In this case, we had able to consolidate more servers into every ESX host, and users do not experience slow performance after the changes been made. As now, the maximum CPU utilization on the ESX host always remain at 70% at the peak hour which is due to the time zone crossing from Asia pacific with EMEA hour. This peak utilization will not stay longer than 4 hours and it will back to normal once Asia Pacific office hour is end.
This trick may not work for everyone, but if your environment do similar with what happen here, this should work. As today, I do not utilize the resource pool after the virtual machines re-allocation I did. Of course, you may need to assign the resources carefully to every virtual machines.
Related posts:
Tags: CPU, ESX, resource pool, virtual machine, VM, VMware
2 Responses
Do you use DRS?
If you do, VC would dynamically re-assign VM’s to hosts based on current utilisation.
yes, I use DRS, but I do not set to the most aggressive level, DRS will only kick in as it needed. Our vm is sitting on R900, therefore, initially, the DRS will not happen too frequent and to have plan proper in the initial stage to prevent too many DRS happen will be better. Is my personal opinion