In this post, we will examine a variety of scenarios you might need to plan for in your datacenter.
The Status Quo
I think I’ve heard all the possible answers. “So how do you do capacity planning today?” The responses usually sound something like:
“The ol’ finger in the air…”
“We’ve got a pretty complex excel sheet.”
“We’ve got a team dedicated to planning. They use quite a few different tools.”
“Psh, capacity planning?! We just found out about a project we were supposed to support yesterday.”
Regardless of the category you find yourself in, you always end up with the same risks. Did I just over spend on hardware? Will I have enough hardware to support our demands? Can my team manage the additional growth and complexity? With the current approaches to capacity planning it’s virtually impossible to be able to manage the tradeoffs between budget, performance, and productivity. Here’s how to change that.
Using Demand to Plan, Not Supply
Instead of looking at the utilization of hardware to plan for growth or changes, look at the demands of your workload. Which resources, and how much, do each of your applications demand in order to operate reliably? Once you know this, all you have to do is align that demand with proper amount of supply (compute, storage, network). Of course, aligning demand to supply continuously and in real-time is far too complex for humans. If you’re not familiar with Turbonomic’s autonomic platform, give this page a quick read to learn more about why understanding demand is so important to making capacity decisions in both real-time and in future planning.
Here we go!
Scenario 1: Projection
The Question: What will my datacenter look like in 3 months (or 6 months or 12 months) based on historical growth?
One of the most basic questions anyone needs to have an answer for when walking into that budgeting meeting. In this video, we’ll start simple but in Scenario 2 we’ll add to this and cover more advanced projections.
Ok, we’ll start off the Planning Scenarios with a Projection. The use case for a plan like this would be to determine how much hardware is needed at a future point in time based on the historical growth of demand for resources. Some things to think about: How far into the future would you like to plan? At what interval would you like to see the data points for the plan (every month, every quarter?)
So here’s how we do a Projection plan very quickly. First click on the scope button and select the cluster or datacenter you’d like to work with. Now let’s turn on Projection. We are now presented with a few options. How far into the future would we like to project? Select your end date. Based on the number of months into the future, Operations Manager displays the divisible monthly intervals for which we can plot the projection. Click run. Here are the results! For each monthly interval we can see the total number of workloads we need to anticipate and the necessary hardware needed to support their demand for resources. That was pretty easy! Let’s save and export to a PDF.
Scenario 2: Projection and Projects
The Question: What will my datacenter look like in X months if…
…I account for previous months’ growth?
…I am planning to consistently add workloads every month?
…have to support a large project in additional to normal growth rates?
So now that we know how to setup a Projection plan, let’s build off of that to accommodate these additional scenarios.
Ok, so after applying our knowledge from the first scenario, let’s setup a new Projection Plan. Again, we’ll scope to a particular cluster and enable Projection. Now let’s consider some new factors this time, like growth observed in previous months and/or repeated growth we anticipate this year. Let’s click into Demand and check out some new settings. Expand “Add VMs” and select “Add VMs based on previous months. When this setting is enabled, the Plan will use the previous month’s data of VMs added to this scope, and then add that same number of VMs each month until your specified end date. For instance if 3 VMs were added last month and you projected 3 months into the future, the plan would add a total of 9 VMs. Next select “Copy VMs”. Outside of the previous months growth, you know you’ll be adding some more workloads because of an upcoming project. You can use the existing VMs and create additional copies which allows Operations Manager to use the information it already has on those workloads. Will these be a set of VMs you are adding just once? Or will you be regularly adding these VMs on a monthly or yearly basis. Use the “When” drop-down to select your preference. I’ll select “Monthy” for this set of VMs. Let’s run. Here are the results! For each monthly interval we can see the total number of workloads we need to anticipate and the necessary hardware needed to support their demand for resources. That was pretty easy! Let’s save and export to a PDF.
Scenario 3: Planning for the Peaks
The Question: What resources are needed to accommodate a peak period AND new workloads?
Understanding the answer this question is extremely helpful for those that support environments with recurring periods of high demand – that could be weekly, monthly, quarterly, or annually. Let’s take a look at using Turbonomic's Historical Baseline capabilities to plan for this scenario.
Ok, running a plan using Historical Baseline & Peaks functionality is a helpful one especially for those in an industry with high seasonal demand. Let’s review an example. To start, let’s set the scope to a particular cluster or datacenter. Next, let’s open up the Baseline menu. Select “Use state from history:” You’re now able to use the slider at the bottom to scroll back in time and select a peak period of utilization. You can toggle the different resources at the top to plan for a specific resource spike. Select the point in the time and click “Ok” and “Done”. Now that we’ve got our Peak baseline set, let’s add some additional workloads to find out what hardware is needed to support new load on top of peak utilization. Let’s open up the Demand panel and add some additional VMs based on existing workload. Now let’s run. We now know what hardware is required to keep the environment in a desired state when running at a historical peak supporting additional workloads. The Workload view in the Plan is another way of visualizing the impact of the decisions at the bottom. Here we see we can eliminate the risks associated with VMs running on heavily utilized storage and hosts, and instead operate in the desired state on the right. We can then save this plan and export to PDF.
So we’ve just reviewed three, seemingly complex planning scenarios that can all be completed in a matter of seconds.
But this post certainly doesn’t cover all the possible plans you might want to run. Think, hardware refresh, HA, disaster recovery…the list goes on. I’ll cover those scenarios in Part 2 of “How to Run Every Capacity Planning Scenario in Seconds.”