“Highload is when normal doesn’t work.” How Ukrainian online services cope with traffic that’s growing like on steroids
How Favbet Tech, Boosteroid and «Lanet» withstand peak loads and prepare for highload in an adult way
How Favbet Tech, Boosteroid and «Lanet» withstand peak loads and prepare for highload in an adult way
What does a company do when tens of thousands of users log into its server at the same time? When marketing suddenly launches a promotion, streaming must display images without delay, and network traffic increases many times over? Those who work with highload say: standard approaches are no longer enough. dev.ua learned how Favbet Tech, Boosteroid, and Laneta deal with this.
Highload is not about numbers, it’s about flexibility
«I would say that a system becomes highload when it can no longer be built using conventional approaches. When you have to be ready for extremes,» says Artem Zhdanov, development manager at Favbet Tech .
And he adds: looking only at the number of requests is wrong. For one product, 6,000 requests per second is a challenge. For another, it’s a normal day.
The key is whether the architecture has the ability to hold peaks.
What does Favbet Tech do during peak times?
Favbet Tech prepares in advance — the system is designed so as not to «put out a fire» in the midst of a big football match or promotional campaign.
Infrastructure:
- They use Kubernetes, which automatically scales depending on the load.
- Load balancing allows you to evenly distribute requests.
- API Gateway processes traffic, is responsible for authentication, and becomes a critical node in the system.
«Kubernetes adds or removes nodes based on metrics. This allows you to process traffic in real time, rather than queuing requests,» says Zhdanov.
Peak scheduling: is it realistic?
If there is an event (match, promotion) — the team calculates the expected load, estimates queues, checks processing time. But sometimes there is a surprise. “ Marketing decided to issue bonuses — and bang! Tens of thousands of users per day. Before that, everything was calm. Such peaks have to be endured.»
The largest loads that had to be supported were half a million users. But it’s not just about the number. It’s important that everyone goes through a complex player journey with several stages, where bonuses move from one state to another according to clear rules. Some operations cannot be parallelized — they must be sequential.
«And although half a million is not a lot, the system must be ready for this load,» says Zhdanov.
What to do if you can’t stand it:
- analyze metrics;
- expand server resources where possible;
- optimize bottlenecks in code and work with databases, caching services, and message brokers;
- work with technical debt on an ongoing basis.
Boosteroid: «The cloud is expensive.» That’s why we’re building our own
Ivan Shvaichenko, CEO of Boosteroid, which allows you to play heavy games directly in the browser, says: infrastructure in the cloud is a financial dead end.
«Renting servers from hyperscalers like AWS, Azure, or Google Cloud becomes financially unprofitable very quickly. Especially if you have a large-scale project with GPUs,» he says.
What they do:
- They are building their own infrastructure — data centers in the USA, Europe, and Latin America.
- They manage thousands of GPU servers.
- According to Shvaichenko, this is 4–6 times cheaper over five years than renting cloud capacity — even taking into account investments.
«This allows us to keep the price no higher than $10 for users. And this is critical for us,» emphasizes Ivan Shvaichenko.
Lanet has its own highload: peak traffic and custom software
For the Internet provider Lanet, highload is evening traffic, when hundreds of thousands of subscribers access streaming, YouTube, and TikTok.
Marian Ivasyuk, the company’s director, says:
- They keep double the capacity reserve for evening peaks.
- They cluster the infrastructure and balance the load on services.
- They optimize the software as much as possible — where necessary, they write their own.
«Usually we try to squeeze the maximum amount of traffic out of each unit of equipment,» says Ivasyuk.
When AI comes into play
Even in highload infrastructure, there is a place for artificial intelligence — it can predict peak loads and reduce the intensity in advance.
«If AI sees that requests are growing, it can reduce the speed of their processing so that the database does not collapse. This is quite possible, even at the Kubernetes level,» says Zhdanov from Favbet.
WebAssembly, quick start and the future of highload
In the future, highload systems will be replaced by WebAssembly, a technology that allows you to run code on any host at maximum speed.
«Some services start in a split second and take up only 30 MB. This is a breakthrough. Amazon already offers launch via ready-made images — you don’t even need to download the container from scratch,» says Zhdanov.
Why engineers should go work with highload
Favbet Tech says: they are not afraid to give engineers freedom and let them work not with routine, but with real engineering tasks.
«If you want to automate, optimize, remove unnecessary things, come to us. We value people who want to build, not just execute,» says Zhdanov.
Those who work with highload often grow with the system — you can’t learn experience in courses, you have to live it.
«Our specialists grew up with the products. It’s not like that — you won’t learn to work with highload if you haven’t seen a real highload,» summarizes Marian Ivasyuk from Lanet.
Highload is when everything is not according to the instructions. And when it still works.