Black Friday 2017 - performance round-up

Another year, and another Black Friday has come and gone. We’re pleased to report that many of our customers sailed through the biggest sales event of the year with little or no impact on performance. In more than one case, load testing in advance of the peak was critical to that success.

Our customers’ peak was our peak too, and our real user monitoring platform was also up to the task, easily coping with many times the usual number of beacons on the day.

Not everyone was so fortunate.

Outages

Several retailers saw outages during the day, and the graphs below are taken from NCC Group’s monitoring service. The following site, for example, was down twice on Black Friday. It also experienced a slowdown during the afternoon, although this was due to a slow response from a third-party service (Peerius), which wouldn’t have impacted user experience.

Black Friday outage

Another retailer was down for three hours the previous evening:

How one retailer failed the night before Black Friday

The graph below shows how one site took longer and longer to respond (illustrated by the blue area in the graph), until it eventually stopped responding altogether:

A Black Friday outage, preceded by a slowdown

The impact of bigger pages at peak

Some retailers contributed to their own slowdowns by delivering bigger web pages. The example below shows how one page doubled in size on Black Friday and saw a corresponding increase in load times. Performance on the day was stable but slow.

A bigger, slower Black Friday home page

This wouldn’t just have impacted performance – there could also have been an increase in bandwidth and CDN costs.

Others did precisely the opposite, delivering lighter, faster pages on Black Friday. This helped both to keep costs down and to reduce the risk of alienating customers.

The following retailer’s home page was smaller on Black Friday than it had been at any time in the previous month. And although load times were a little more variable on the day than they were before, they were still faster:

Faster load times on Black Friday

Third-party performance

One thing retailers can’t control is the performance of the various third parties they use. For the most part, the best they can do is minimise any impact on the user experience.

We saw a number of third-party slowdowns and failures over the Black Friday period. The good news is that they didn’t appear to cause major problems for their host sites. For example, Sessioncam failed to load in tests for both the retailers below (the first at 5pm and the second at 6pm):

Third-party slowdown at 5pm

Third-party slowdown at 6pm

Supporting bricks and mortar retail

We also saw the impact of Black Friday on bricks and mortar retailers, with one of our customers delivering supporting services for retailers in-store.

Real user monitoring data for that customer shows that there was a big spike in traffic on Thursday evening and again during Friday. Although load times increased, they remained at around the 3 second mark – no different from a typical day. This was great news for our customer – and our customer’s customers.

No additional slowdown despite the extra traffic

Queueing

Several retailers implemented queuing systems this year, while a few others probably wish they had. Of course, a queue is no substitute for a working website, and it’s better to treat it as the last line of defence than the first. One of our customers, for example, had a queuing service waiting in the wings just in case, but didn’t need to trigger it all day. Another retailer, however, generated a furious response on social media as visitors queued for prolonged periods at positions well over 1,200. 

Summary

Overall, the big players seemed to cope pretty well, and we’re pleased to have played a part in that through load testing and performance optimisation. Anecdotally, we seemed to see more failures among smaller, high-end retailers. Less accustomed to very large volumes of traffic, they probably experienced bigger proportional rises than their more mass-market counterparts.

That said, there were still a couple of high-profile outages, proving that no one can afford to be complacent, and no one is too big to fail.

Published date:  28 November 2017

Written by:  Alex Painter

comments powered by Disqus

Filter By Service

Filter By Date