Third Time’s a Charm? More Like Fifth Time
by Shannon Loomis, PhD – Data Scientist
One of the many challenges facing data center architects and operators is the constant balancing act between performance and risk. With storage systems, for example, you want to upgrade to the latest version of your arrays’ operating system to get the new features – like and security enhancements – but you don’t want to get stuck with problems if there are bugs.
Knowing when other customers have upgraded to a new operating system could help you make an informed decision on when to upgrade, but usually this type of data isn’t even measured, let alone provided as a benchmark.
Nimble Storage is different; every Nimble array comes with the InfoSight predictive analytics system built-in, which provides customers with unprecedented insight into their own storage systems. The data from 7,500+ customers are then anonymized and aggregated into a massive data set. This enables our team of data scientists to extract some very useful patterns, including the rate at which customers upgrade to new releases of the Nimble operating system (NOS).
Three major trends emerged when I used InfoSight data to map out NOS releases:
- All upgrades follow similar patterns;
- Most customers don’t upgrade immediately;
- Upgrade rates increase with every feature.
We’ll look at each of these points in more depth to help you decide when an operating system upgrade is right for you.
All Upgrades Follow Similar Patterns
The most obvious trend is that after each major Nimble operating system release, customers start adopting the new OS at a fairly steady rate. The adoption of the new NOS continues until the release of the next upgrade in that major release (e.g. NOS 2.0, NOS 2.1, etc.), after which the percent of arrays across the installed base on that NOS starts to slowly decline. The exceptions to this are the last releases within a major release, which show an earlier decline due to the availability of the next major release (e.g. 2.2.11 saw an early decline due to releases in the 2.3 series).
You can see this pattern in the charts below, which plot the number of arrays (as a percentage of the installed base) against the number of days since the release.
Percent of arrays on a particular release plotted against the number of days since the release. The percent of arrays is normalized such that the maximum is 1, and the number of days since release is normalized such that the next release date is 1.
Most Customers Don’t Upgrade Immediately
Given that data storage is central to every company’s existence, it comes as no surprise that most customers don’t upgrade the moment a new major release becomes available. On average, the most widely installed NOS at any one time tends to be somewhere between the 4th and 6th maintenance release within a major release. The first two releases tend to only be used by early adopters, while the last few releases tend to only be used by a few hangers on. In other words, it’s the fifth time that’s the charm, as demonstrated by the graph below on the left showing the maximum popularity of maintenance releases as a function of their iteration within a major release cycle.
Furthermore, we see that just before a major release, the installed base tends to have an equal mix of the most recent and the penultimate release group (e.g. just before our 3.0 release, there were roughly equal amounts of 2.3 and 2.2 NOSs throughout the installed base). This is displayed in the pie charts below on the right, which show the NOS distributions across the installed base just before a major release.
Left: Maximum percent of arrays against the release number within the major release. Note that the most popular NOSs tend to be between the 4th and 6th maintenance release.
Right: Pie charts showing the NOS breakdown across the installed base just prior to a major release.
Upgrade Rates Increase with Every Major Release
One striking trend is that with every NOS major release, the maximum upgrade rate within that release increases. Furthermore, the upgrade rate continues to increase with every maintenance release in its group until a new major NOS is released. The increasingly higher upgrade rates with each major release demonstrate the continued faith that customers have in Nimble products.
Both of these points are featured below in the plots showing the initial upgrade rate against the number of the release (left) as well as the release date (right).
Left: Initial upgrade rate of all released NOSs shown in percent of arrays upgrading per day. Note the increasing maximum adoption rate with every major release.
Right: Same data as left, but plotted against release date. Note the drop off in upgrade rates after the subsequent major release.
In fact, Nimble customers trust the reliability of NOS so much that a survey last year showed that nearly two-thirds of them upgraded their system’s OS during peak business hours, which is an astonishing statistic.
I’ll leave you with one final piece of data worth noting: around 200 Nimble customers (~3%) are still using a fairly old version of NOS (1.4 or 2.0), which means they’re missing out on all kinds of features. If you’re among these customers, go ahead and upgrade: thousands of others have proven that it really is risk-free.