Prism Blog

Archive for the 'Google Analytics' Category

Back in the academic groove with Google Analytics

We started using Google Analytics with Talis Prism 3 back in March, when usage across tenancies was relatively low – in the hundreds. Of course, in the intervening period a significant number of customers have adopted Talis Prism 3. As Imraz Mohammed pointed out on his recent Talis Prism 3 Update on this blog, 30 libraries have now moved to Talis Prism 3, of which 13 are university libraries. The combination of these increased volumes of usage and the approach of the new academic year means that Google Analytics is proving invaluable to Talis development staff and customers alike. Not only does it provide an updated picture of usage every few hours, but it’s so flexible that in different contexts, in this case, different times of year, it can provide powerful metrics for diagnostics and decision-making.

Of course the start of the academic year is particularly interesting (or murderous, depending on your viewpoint) for university libraries, where the user base and its multifarious motivations are shifting almost by the day. Around the middle of September, we were running at around 9,500 unique visitors per day across all the Talis Prism 3 tenancies. At this point it can be useful to start using the map facility on Google Analytics to get a picture (literally) of the shifting balance between use on campus and access from students at home.

And then suddenly, from Monday 27th September, universities welcome the year’s new students and the usage statistics started to climb. University of Northampton provides a great example of this. Already by Monday 27th, Talis Prism 3 is handling twice the previous traffic. On Tuesday there were 39,000 and by Friday they had reached 59,000.

By week beginning  4th October, students are returning, but freshers continue to make their presence felt on Google Analytics. The important metric Average pages per visit decreases as freshers log on to take a look at the system and play around with a few cursory searches before lectures start. On a similar vein, Average time spent on page has dipped and the bounce rate is up.

We start to see patterns at many levels. We notice that usage peaks on Wednesday and Thursday and then calms down for the weekend, before a growing again from Monday. And the longer we use it the more useful those patterns will be. I look forward to next year when we’ll be able to refer back to the September / October metrics in order to anticipate usage.

Our interest in Google Analytics extends well beyond the value for academic libraries in understanding end-user behaviour, important though that is. When providing Software as a Service such as Talis Prism 3, we need to be able to anticipate load with a high degree of accuracy. If not, unexpected spikes in usage may compromise the service. And this of course is the most critical time of the year, when expectations among students of campus services such as library systems are most likely to be set. On 27th September, when we experienced 1.5 hours of outage on Talis Prism 3, it was crucial to diagnose, resolve, and also make sure the problem did not recur. Two weeks later, Google Analytics enables us to state with confidence that we are successfully handling a five-fold increase in traffic compared to 27th September.

In order to resolve the underlying problems behind the outage, we advanced some of the performance enhancements we’d been working on into production. The majority of these changes will be released shortly, following a series of phased roll-outs, in which we install change on half the servers, monitor, and then roll out to the rest of the pool. This ensures that changes are tested in isolation, and is part of a broader testing strategy around Talis Prism 3, centred around:

a.       Automated tests which, for every change introduced, run through every Talis Prism 3 transaction – e.g. renewal, search, login – checking for an unchanged output in every instance. 20-30 minutes is all it takes to run through this comprehensive suite of functions.

b.      Performance tests in which we mimic live traffic on the two staging servers here at Talis before release onto customer tenancies.