The remaining capability on which is not sufficient add a new VM
As companies seek to gain speed, agility, and cost savings out of their virtualization initiatives, trying to improve the utilization of their resources can be a difficult task. Typically, only about 25 percent of the available processing power of virtualized servers is being utilized, where those utilization numbers should be up around 55 to 60 percent to gain the true economies of running virtualized applications (see CIO blog interview with Gartner analyst David Cappuccio).
IT organizations need analytic capabilities to help them optimize their virtualized environments and to help them make intelligence decisions when it comes to capacity planning for the future. Shiva Prakash from HP’s Service Intelligence R&D team explains some of the tips and tricks for making optimization an easier task.
In spite of IT organizations adopting virtualization, the average utilization of the data center resources remains fairly low. Using tools for rightsizing the VMs and achieving optimal placement of VMs to minimize the holes on the hosts have helped administrators to achieve some improvement in the average utilization, but this improvement is far less optimal than original expectations from virtualization.
The secret to achieve higher utilization in virtualized datacenters is not just consolidating resources, but doing consolidation with ‘optimal oversubscription’ of resources. Optimal oversubscription can be achieved by analyzing the trend of resource usage by VMs and maximizing the percentage of over- subscribed capacity by co-locating VMs with complementary resource usage trends. In VMware, the higher density of VMs also increases the probability of further optimizing memory usage through TPS.
Let’s look the example illustrated below…
In the above example, both resources on host A and host B are optimally allocated (roughly 85%), but the average utilization is very low. If you look at the resource usage trend of OLTP workloads on host A, you will see that these VMs are very active during business hours compared to non-business hours. Similarly, DWH workloads on host B are very active during non-business hours compared to business hours. It is clear that by having a mix of OLTP and DWH workloads on the same host, we can actually over subscribe the host resources and place a higher number of VMs without making any of the VMs starve for resources, and hence achieve the higher average resource utilization of Hosts.
The key in making this approach work is the existence of workloads with complimentary resource usage trend patterns. The probability of existence of such workloads is higher in medium to large size environments.
In the above example, I have simplified the scenario drastically to explain the concept but in real life, it is impossible to do this kind of analysis without help from analytic modeling tools when you have hundreds of VMs coupled with practical constraints. Products like HP Service Health Optimizer will help you in identifying optimal VM placement by automating the analysis of seasonality patterns in resource usage, and identifying the complementary loads for best placement for targeted Virtualization technologies.
Product Marketing Manager for HP Application Performance Management suite of software products. Before this role, I worked in the HP StorageWorks Division working as both a Product Marketing Manager overseeing enterprise hardware and software, as well as working as Business Development Manager for the Enterprise Services channel.