Hello Friends!  Welcome to another installment on Network Capacity Planning with SevOne! 

Today we are covering percentiles -- what they are, how to calculate them, how to use them in capacity planning efforts.

Introduction

When planning for network capacity, your goal is to gain an understanding of how your current bandwidth is being used by the business.  Usually percent (%) utilization is used in order to normalize how you review bandwidth usage for a set of interfaces regardless of the amount of bandwidth actually allocated to each individual interface in bits per second (bits/sec).  With time-series data, IBM's SevOne NMS typically collects data from interfaces at a 5-minute poll rate.  In some cases, the network engineering team may choose to poll certain interfaces more frequently, using a 1-minute or 30-second poll rate.  In either case, you end up with thousands of poll values to analyze when it comes time to assess capacity needs.  Here's where percentiles come in handy.

Percentiles - What are they?

When planning for network capacity, the goal is simple -- make sure you have adequate capacity available for your network links so you don't impact end-user productivity and user experience.  Usually, WAN links are where the lack of available bandwidth can become an issue.  When viewing a trend chart showing bandwidth utilization, you can see peaks and valleys over the course of days, weeks, months, which make it visually possible to identify max and min values, but it is hard to understand what percentage of the time your bandwidth is being highly utilized, maybe nearing capacity where end-users might start to be impacted.  Network capacity planners like to view percentiles when analyzing capacity needs.  Percentiles tell the capacity planner what percentage of samples collected reached a network capacity of 'x %' or less during a time period.  So, the 90th percentile would show what level bandwidth reached for 90% of all poll values for the period (e.g. 90% of polls were at 79% or less).  95th percentile would show the maximum bandwidth that was reached for 95% of all poll values (e.g. 95% of polls were at 87% or less).  

How are Percentiles calculated?

Here's how percentiles are calculated:

1. The first step is to take all of the polled values and place them in an ordered list (array). 
2. Then sort the values ascending. 
3. Determine the maximum number of samples in your array of values.  If you are analyzing network capacity for the past 90 days, using a 5-minute poll rate, you would be analyzing 90 days x 24 hours x 12 polls per hour = 25,920 samples. 
4. To calculate the 95th percentile value, you would multiply .95 x the number of samples, which would provide you with the index number in the array associated with the 95th percentile. 
5. Retrieve the array value using the index number to identify the 95th percentile value.  So, for our 90 day example, .95 * 25,920 (samples) = 24,624.  So our 95th percentile value would be found in our array at index 24624.  If the value found there is 82.5%, then you would know that 95% of all samples were at 82.5% utilization or less.

How are Percentiles used in capacity planning?

Percentiles provide a quick and accurate snapshot of the maximum bandwidth percentage that was consumed for 'X' percent of the samples.  If you are at 90% utilization over 95% of the time (95th percentile), and are at or near capacity when you account for 99% of all samples (99th percentile), then it's probably time to upgrade a circuit's bandwidth capacity.  Depending on your company's tolerance for network saturation, you may choose to upgrade your bandwidth more aggressively (sooner) if you cannot tolerate any saturation (e.g. financial, stock markets use cases, etc).  Usually capacity planners will view 90th, 95th, 99th, and max values to understand whether an upgrade is warranted or not.

What best practices should be used when using percentiles during capacity planning?

• You need all of the data poll values (samples) for the time period you are considering in order to calculate an percentile index number so you can locate the percentile value accurately.  IBM SevOne keeps all data poll values for one year without rolling them up allowing you to view accurate percentile values for the entire year. 
• When planning for capacity, you should exclude non-business hours polls so that your result set contains the polls that really matter to the business.  IBM SevOne permits you to calculate percentiles using either all polls or just the business-hours polls, providing you with an accurate result no matter your situation.  

Where are percentiles found within SevOne reports?

TopN Views that contain percentile values can be displayed in the IBM SevOne Data Insight reports using the TopN widget.  Percentiles values can be added to any existing TopN View by making a copy of it and then adding the values.  This is done in the TopN view editor on the IBM SevOne NMS back-end system.

You can also display a percentile value as a dashed line within a Performance Metrics widget', allowing you to quickly see when values cross the 95th percentile.

I hope that this article gave you a better understanding of what percentiles are and how they are used within IBM SevOne by capacity planners to make capacity decisions.

#TechnicalBlog