Articles Monitoring Capacity and the 4th Dimension
Capacity and the 4th Dimension
Friday, September 16 2011 14:30 Written by VMGuru
I was sketching out some diagrams today and started to think about how the time frame that we analyze and how it relates to Capacity Management. The concept of "Working" or "Business" hours actually plays a vital role in how capacity is accounted for and accommodated in the datacenter.
I often relate introducing business hours into capacity management to being a decision point during a breakup. There is a line that is drawn between "Lets remain friends" and "Never call me again". With business hours, we need to determine how big of a window we want to consider for the data we care about. This is going to be different for every organization out there. For the purpose of simplicity, let's look at a window in which I only care about data between 6am and 6pm.
Why is a time window important? I have found that many organizations really don't care about what happens during off-hours time windows. The data they really care about is "How hard are my systems working when people are using them". Let's take a quick look at how analyzing off-hours data can impact the data we use to provide Capacity Projections about our environment.
In the previous image above, I've drawn in 3 average lines. The Green Average line is the one we REALLY care about. It dictates how the systems are being utilized when people are actually accessing them. The Pink Average indicates the average utilization of systems when no one is around the office and they aren't doing anything. The Blue Average is the total average across the 24 hour time window. This is skewed to include statistics from when systems both are and are not being accessed.
If I base my capacity calculations off the Blue Average, I get results that do not jive with reality. It's skewed to include utilization valleys from when my systems aren't being accessed. What I really want to consider is the Green Average, which is the metric of what my systems are doing while they are being accessed. The difference between the Green and Blue averages could have a profound impact on determining how much more capacity is available based on traditional workload statistics.
The Pink Average doesn't need to only consider utilization from when systems aren't being accessed. Let's assume that most organizations also have a backup process that starts at 10PM nightly. We know this spikes the environment for 3 hours at 100% utilization. This is known and expected behavior, that could also have an impact on the average utilization and inaccurately skew results. Since this backup traffic is predictable and controlled, we don't want it getting in the way of real data, so don't want it included as "Business Hours" data.
I'm sure there are several scenarios that people can think of (Virus scans, etc). It all goes to highlight how considering business hours is vital to providing an accurate assessment of your virtualization capacity requirements.
TrackBack URI for this entry
Show/hide comment form
| Next > |
|---|