Something very exciting has happened.
We’ve just put our load testing platform under more strain than ever before.
The previous record was set back in 2011, when we hit a site with 457,000 page impressions per minute.
This time we reached a staggering 868,000.
Now, more astute readers will note that this is somewhat short of the million quoted in the headline (in a shameless ploy to get you to read on).
However, we think we can say with some confidence that the million is well within our grasp. Here’s why:
This graph shows the CPU utilisation for a single load injector during the test. It hit a maximum of 20%. Since this particular load test had no central dependencies (such as incremental datasets), this can be scaled fairly well across all injectors. Giving the system itself some breathing room, we can say that this load test used about 25% of the total available CPU resource.
This graph shows the memory used during the test: a maximum of 22.7GB out of a possible 188.9GB. Again, if we give the system some leeway, this translates to about 15% of the available memory.
Finally, here’s the bandwidth used on one of our four connectivity drops during the test. Each drop is 1Gb/s, and the maximum ingress data rate we saw was 41.2MB/s. After allowing a bit of latitude, this amounts to about 40% of the available bandwidth.
In summary, although this load test more ambitious than what we’d done before, we’re confident that we can cope with much more. For those of us who have been here since 2011, it also highlights just how far we’ve come in that time.
Of course, we’re not saying that we could manage a million page impressions a minute under any circumstances. Every job is different. But we put a lot of time and resources into building and maintaining a stable, resilient platform, and it’s great to see that platform cope under pressure.
Owen is a Senior Systems Administrator at NCC Group Web Performance.