Valve has 1,700 CPUs working non-stop to prevent hacking in CS:GO

All the popular multiplayer games are joining the never-ending battle against hackers. For Counter-Strike: Global Offensive, the most played FPS game in the world, preventing CS:GO hacking is always a top priority for Valve and players.

As the 10th game released on Source (and the third part of the mainstream CS series), the amount of knowledge about this platform is already known to many people, so they can ‘play’ with Valve’s engine. Hacks created in the aging game Half-Life 2: Deathmatch, which only takes a few minutes to edit, can work well in CS: GO (but Valve says they are too easy to detect). In terms of design, the qualities of CS:GO as a game leaning towards accuracy as well as skill have made hacking more rampant. The weapons can all be ‘one shot one kill’, making the aimbot more terrifying than ever. CS:GO’s focus on information and stealth makes it extremely important to know where your opponents are – a breeding ground for widespread wallhacking.

According to Vavle, the fight against hackers in CS: GO is “an extremely important and precious work”. But if you have played this game, you will have noticed that: things are gradually improving. Complaints and hacking clips on the Reddit forum not only appear less, but concerns about “hacking like going to the market” also gradually disappear. Valve has tried to ban thousands of hackers in recent VACs. So how did Valve deal with these guys?

Valve programmer John McDonal spoke at GDC last week in San Francisco about what in-depth techniques he and Vavle have used to tackle hacking in CS:GO. This method has become so effective that Valve is now using it to apply it to “so many other things”, from anti-fraud to several aspects of Dota 2, and Valve is working on it. Working hard with other studios to find anti-cheat solutions for other games on Steam.

Solve hacking problem in CS:GO

At the time of 2016, McDonald’s noticed “The only thing that the community talks about is all about hacking,”. This information comes from online chats and private email addresses received from CS:GO pros. The increase in VAC bans during this period clearly shows that the hack is slowly being abused.

To combat this problem, Valve and McDonald’s were looking for a solution that would work on its own and adapt over time to new hacking techniques. This is a quality of Valve, they have a history of automating the way they operate on Steam, instead of having to hire hundreds of new employees to manage. Valve created VACnet, a project that took a year to develop.

VACnet has partnered with Overwatch, a CS:GO player replay viewing tool, to beat players reported for their bad behavior. VACnet is not just a new form of VAC (VAC uses a method of parallelizing the server and the software in your computer, for example: the user runs the hacking software along with the game). VACnet is brand new system added. It analyzes user behavior in the game, to identify what hacking is like, and then finds and bans hackers from the learned criteria.

McDonal said that “closed” hacks are still difficult to solve, but when building VACnet, Valve decided to target aimbot first. Because they are easy to spot in certain rounds of CS:GO: when fired. This allowed Valve to create a system that detects unusual changes in pitch (vertical) and yaw (horizontal – a measure of the amount in the player’s screen – for half a second before a shot is fired, and A quarter of a second later, this data, along with other information like what weapon they’re using, the distance, and the result of that shot (hit, miss, headshot?) will combine. to create “atom” (as Valve coined it), which is a package of data to describe each shot.

However, VACnet cannot detect a hacker based on just one atom.” We need an atom string, what we want is 140 atoms, or at least what the system is using right now… I just took 140 shots out of 8 rounds, put them in VACnet and watched ‘Hey, if we give a 140-shot sequence to a ‘hit human’, what’s the probability that that person will be deemed hacked? ”

The results are quite positive. Both player reports and VACnet are used for judgment in Overwatch. But when VACnet suspects a hacker, it’s almost certainly a hacker.

“When a person submits the results of their review on Overwatch (of a suspected player), the chance of them being convicted is only 15-30 percent, and this number depends on many factors. factors, like the time of year, is the game on sale, is it spring break. There are many reasons, but the most important thing is that the player’s assessment of it as a violator is very low,” McDonald said. “The conclusion of VACnet has a very high probability, when VACnet reports it is usually 80 to 95%.”

This doesn’t mean Valve will get rid of the Overwatch array. Both systems worked: VACnet discovered new techniques from Overwatch. McDonal said. “Because we use Overwatch and Valve doesn’t really replace all player reports, we just add support, which means VACnet will improve with player reports. When players discover new hacking methods, VACnet will learn too.”

McDonald added that when VACnet was retrained with data from users to detect new hacks, the accuracy rate reached nearly 100% shortly before the hackers further edited. When Valve quietly brought VACnet to CS:GO in 2v2 mode last week, “the conclusion accuracy rate in that mode hit 99% for a while, which is great. The hackers didn’t know what we did, and the players were happy about it and Valve swept the hackers. So it feels really good.”

Giant anti-hack accelerator

To make VACnet a reality, a server system had to be built that could handle millions of CS:GO players and tons of data while the game was getting more and more players. Currently there are about 600,000 5v5 CS:GO matches per day, and to evaluate all the players in those matches, Valve needs about four minutes of data processing, or 2.4 million minutes if only one CPU is to work. that everyday. You need up to 1,700 CPUs to process this information on a daily basis.

So Valve bought 1,700 CPUs. And 1,700 more CPUs, “so we can develop more,” McDonald said, indicating Valve plans to include VACnet in other games as well. Overall, Valve spent at least a few million dollars on this system: 64 server bars with 54 CPU cores each and 128GB of RAM in each. This number is a small drop compared to the $120 million CS:GO made from non-game sales in 2017, but it is clearly one of the strongest anti-hack machines for a single game. .

The work isn’t over yet, but from McDonald’s perspective, VACnet is working well, and it doesn’t just apply to non-Valve games, but possibly other things on Steam as well. “Continuous learning technology is a revolutionary approach,” says McDonald confidently. “We think it really helps developers to not be distracted so that it doesn’t affect the customer. Our customers now experience fewer hacks than before, and discussions about hackers have quieted down.”

At the beginning of the month of 2017 a new milestone for this system was reached: VACnet began to achieve a conclusion accuracy rate of over 50% in Overwatch. “The system works great,” McDonald concluded.