Dynamic Optimization of the Private Cloud Infrastructure

Dynamic Optimization of the Private Cloud Infrastructure

  • Comments 5
  • Likes

As more and more organizations are moving towards the private cloud, you may be wondering what you can do to give your apps the best chance of running most efficiently. You are possibly working to create your private cloud. Perhaps, you have it already deployed and configured. Now, you want to help ensure that the Apps running on your cloud are getting the resources they need, when they need them.

With the private cloud, many of the apps running are probably going to have dynamic resource needs (very active at certain times of the day/week/month) and variable activity most of the time. With this, you want your Private Cloud Infrastructure to be able to accommodate these changes in resources by allocating the applications across all of your Infrastructure Components as necessary to handle the load.

With System Center 2012, we have introduced Dynamic Optimization which will allow an administrator to specify thresholds on how loaded or out of balance your individual boxes will get before we rebalance the workloads. With Virtual Machine Manager, a component of System Center 2012, you can set thresholds, frequency, and aggressiveness in how we handle dynamically optimizing workloads within a host group. Dynamic Optimization is set at the host group level, and will be considered for clusters of either Microsoft Hyper-V, VMware ESX(i), and Citrix XenServer Hypervisors.

clip_image002

Thresholds

Thresholds specify the values in percentages or distinct units to initiate an optimization event. On every frequency interval, we will check if the remaining resources for a host in a cluster have fallen below the threshold. If so, then an optimization will occur. When monitoring whether we have surpassed the Dynamic Optimization threshold, we look at a conservative average of the last 10 minutes, which is basically the average of the top 40% of the last ten, 1-minute averages. At this point, we will determine if the threshold has been surpassed, and if it has, will perform the optimizations which will best reallocate the workload within the cluster.

Frequency

The frequency value specifies how often we will perform automatic dynamic optimizations. You will want to base this value on how out of balance your machines are with how often you are migrating VMs between nodes on your cluster. We default to 10 minutes, but you can drop it down or increase it depending on your needs.

Aggressiveness

The value of aggressiveness determines how many VMs will be migrated and where they will be migrated to when an optimization event occurs. When Aggressiveness is set to Medium, Medium-High, or High, we will migrate VMs around if doing so provides a benefit to the hosts (even if the systems are not loaded over the threshold). As long as there is an overall benefit to the cluster, this will happen. Now, the higher the Aggressiveness setting, the more we will migrate VMs around to find the correct balance, and we do this with “Star” ratings. Star ratings are values determined by Intelligent Placement that take into account how many resources are available on a system as well as what the system is doing, and some other things. If we can get a 0.2 improvement in the Star ratings when Aggressiveness is set to High, 0.4 for Medium-High, and 0.6 for Medium, we will migrate VMs around.

As you can see, we try to give the services and VMs you have deployed the best chance at running well by doing our best to keep the resource loads balanced within the clusters. This will also help handle those dynamically changing resource needs due to known and unforeseen changes in demand. Once you have set this piece up, you should look at Power Optimization for those times when you don’t need all of your servers powered on. Then, when you are ready, check out the reporting capabilities that will not only allow you to see how the resources are being used, now, but trend their usage over time, and help to predict the needs of the future.

Kenon Owens

System Center Technical Product Manager

Your comment has been posted.   Close
Thank you, your comment requires moderation so it may take a while to appear.   Close
Leave a Comment
  • Post got lost... Here it is. Again:

    Why does this only work with clusters? Live migration is possible without clusters after all.

  • So if the aggressiveness  values range from 0.2-0.6 for high to medium is it fair to assume medium-low would have a value of 0.8 and low would have a 1 star requirement?

    Paul

  • Q: Why does this only work with clusters? Live migration is possible without clusters after all.

    A: Though there are some reasons why you may want to Live Migrate between machines that are not necessarily within the same cluster, Dynamic Optimization was designed to Optimize at the cluster level because we allow multiple different Hypervisors and clusters to be within a Host Group and clusters gives us the logical grouping we need.

  • Q: So if the aggressiveness  values range from 0.2-0.6 for high to medium is it fair to assume medium-low would have a value of 0.8 and low would have a 1 star requirement?

    A: That is correct that medium-low is 0.8, and low is 1 star, but in the medium-low and low cases, we will only Optimize if you surpass a threshold, not just to balance the load.

  • Can you do, for example, power minimization with dynamic optimization or does it only do load balancing?