Even though that blog post ended up being longer than my usual posts, there’s still additional greatness coming from DRS. I want to focus on Predictive DRS, what it is, how it works, and when you can use it.
Predictive DRS is part of vSphere 6.5. However, it will be in Tech Preview until vRealize Operations releases their version of vROPs which will be compatible with vSphere 6.5, hopefully sometime in the first half of next year. At that point, Predictive-DRS will be fully GA and supported in production environments. From the vSphere side of the house, the code is in place and the feature would be GA, however as many of you know, not all of our products run on the same release cadence. In this situation, we need to wait for the corresponding version of vROps to be released as well.
That being said, let’s talk about what it is and why it’s important. Predictive DRS leverages a tight integration with vRealize Operations Manager (vROps). In doing so, vROPs computes and forecasts VM CPU and memory utilization based on metric history that it pulls from vCenter. vROPs creates Dynamic Thresholds and as more and more data points are introduced over time, the forecasts become more and more accurate with higher levels of confidence.
These forecasted metrics are then sent to DRS and DRS ingests these ahead of time (default is 60 minutes ahead of time) and balances the cluster based on forecasted utilization. The benefit to this is that predictable workload utilization spikes can be acknowledged and balanced before the spike occurs, leading to better performance for all workloads.
So, as you can see in the picture below, DRS in and of itself (Blue line) will react and attempt to load balance when it observes a resource demand spike. But Predictive DRS will have the ability to balance before the observed spike if it has enough predictable workloads. This is pretty awesome! Granted, this will not mean that predictive DRS will get rid of all regular DRS moves, that’s just not realistic. However, over time you should see your regular DRS moves become less and less (for predictable VMs) and potential contention become pretty much non-existent for those workloads.
In vROPs I was able to pull up this active memory monitor for my VM. You can see that every day between 11am-1pm, the Active Memory usage spikes. Because of these consistencies, pDRS has now forecasted when this will occur again and will send its forecasted utilization metrics to DRS again at roughly 10am to prepare and potentially vMotion some workloads around to accommodate before 11am hits.
From the vCenter side of the house, enabling Predictive DRS is as simple as going into cluster settings > DRS > Predictive DRS, and checking the checkbox. I’ll post more on the simple configuration of vROPs side as soon as the vROPs GA for this gets closer.
So, what do you think?