Join the demo

Behind the scenes of our 10k deathmatch

Mark Wright
Apr 2, 2019 7:19:47 PM

There are some more technically focussed blogs coming up in the next few weeks, but we thought it would also be interesting to take a look at the more human element of our 10k stress test.

Why 10,000 players?

When we started looking at what we wanted to do for the Game Developers Conference in the second half to 2018, we knew our friends at CCP already had a platform that their capsuleers had managed to set a world record on - 6,142 players all battled together in January 2018.

The thought popped into our heads “wouldn’t it be cool to break a record at GDC ‘19 as a way to announce Aether Engine”. We could have simply gone for 6,500 players, but if you’re going to try and break a record, you should really go for it, right?

The next round number from that was 10,000 players and we were pretty confident that our technology could handle that many people if we could get them all to turn up at the same time and player for a while.

How did you find the people?

At first we looked at lots of options. We knew that we needed to repeatedly test the whole system, so we wrote some pretty simple ‘bots’ that we could easily replicate. They connected from outside of our Microsoft Azure infrastructure, and we had them collect a bunch of metrics for us, so their ‘experience’ was pretty similar to what real world players should experience. It gave us a sure way to get a very broad geographic spread (more on that later) and push our available bandwidth to its limits. With that done it let us artificially stress the system whenever we wanted and gave us a platform that could gather metrics and let us tweak their behaviour to be more and more like real players.

With our ‘bot’ fleet sorted, we knew we needed real humans to get involved. For one, we wouldn’t be breaking any records with just bots, and secondly, we wanted to see what emergent behaviour developed once real live people were playing around. Our bots will do what they’re told to do - humans, game testers in particular, will do everything they can to stress a system in ways you’d never think of - we had players turn their backs on the fight going on and try to fly away to see how far they’d get; or to see if they could get into one of the titan-class ships scattering the volume.

So we started advertising on social media channels that we wanted 10,000 people to try out our tech demo. We looked at the results of our adverts and we optimised. And then we looked again and optimised again. Ad nauseum. There’s probably a blog post just there.

We started to build some good momentum and realised that we should start explaining what we were hoping to do, when, why, etc., so we got our Discord server up and running, told our mailing list, and realised that we’d started to build a bit of a community. People who were really interested in space games, like EVE Online, Star Citizen, Dual Universe, and people that were genuinely interested in our technology and what it might be able to achieve.

Then Hilmar and our friends at CCP put out the word and overnight our signups doubled. Literally. A day or two later and they’d tripled, until in the end we had over 15k people sign up and register their interest.

Our Discord server went from a few hundred people registered to a couple of thousand people online!

How did we decide when to do the test?

As people were signing up for the test we were looking at which time-zones they were in. On a weekly basis we’d run a report and see where everyone that was signed up was distributed.

We had people signed up from 25 different time-zones, from UTC-8 to UTC+13. Hang on, I hear you say, surely there are only 24 time-zones! Well we had people sign up from Iran, India and Adelaide, Australia, each of which is in a ‘half hour’ time-zone - +3.5, +5.5 and +10.5 respectively

The biggest bulk of our signups was from UTC to UTC+3, i.e. Europe, with the next biggest being the United States, particularly UTC-5 and UTC-6, i.e. the east coast and midwest. On that basis, and because we had the opening talk of GDC, we decided that 17:30 UTC was the ideal time for us to do the test; not too late for eastern Europe and beyond, not too early for the west coast of the United States.

Ideally we’d have run the test during a weekend, to really maximise the number of people that could have taken part, but the Game Developers Conference is strictly a week day affair, and so it made sense to put it between our opening talk of the conference and Hilmar Péturrson’s, the CEO of CCP Games, lightning talk on our stand. If things went as well as we hoped they would (and they did), that gave us the maximum amount of time to talk to people and answer questions about the test for the rest of GDC.

Of course the best laid plans...etc., meant that we actually started the test 30 minutes later than we’d planned.

Where were people?

One of the things we’ve been poor in highlighting previously is that we had a truly global audience for the stress test. Overall we had connections from at least 87 countries. This map shows people connecting just after the test started and you can see that we had players from Hawaii to Tokyo, Reykjavik to Cape Town, London to Wellington, all playing in real time.


Perhaps unsurprisingly, given the fairly heavy involvement of EVE’s capsuleers, our most popular locations were the United States, the United Kingdom, Russia, Germany and Canada - all locations where EVE Online is very popular.

The following video shows the location of people for the first 27 minutes of the stress test.


The one stat we know you’ve all been waiting for...

Whilst the purpose of the test was very much to stress our technology, there were obviously kill/death stats in the client. We didn’t include a leaderboard because of time constraints, despite wanting to, however we were monitoring the numbers.

We can now reveal that the ‘winner’, with a K/D 100/8, was a Brit, with player ID 1851, a.k.a. GladezZ on Discord. Congratulations, GladezZ!

GladezZ fired a total of 1,899 torpedoes, but was nowhere near the trigger happiest of you. That honour goes to player ID 1803, from Michigan, who managed to fire a whopping 2,409. 

You May Also Like

These Stories on Gaming

Subscribe by Email

No Comments Yet

Let us know what you think