Report – Project L Will Have the Most Advanced Network Infrastructure Ever for a Fighting Game

When Tom Cannon/Senior Director and Executive Producer provided a big update on Project L on November 20, 2021 he touched on a big pain point for fighting games particularly in the post-COVID-19 era –netcode.  While Tony Canon/Technical Lead provided us some details on what to expect from the network, an analysis of a session given by David Press/Senior Principal Software Engineer at Riot and Ashwin Raghuraman/Solutions Architect at AWS at re:Invent 2021 about their Valorant deployment provides some insight in to why Project L will have the most advanced infrastructure of any fighting game ever.

Valorant launched with a goal of having 90% of all players in a region under 35ms. After the private beta in April of 2020, they realized that their existing infrastructure in place on Amazon Web Services (AWS) for League of Legends would not be adequate to achieve these goals…so they set about to essentially build a new network, one that could support the low latency requirements of a FPS.

To accomplish the two goals above they set out a plan to improve their infrastructure by moving game servers closer to players and improving routing to game servers. They weren’t content on pursuing a singular options, so they incorporated several very different ones into an overarching strategy.
AWS is market leader in cloud technology and has a presence in regions all over the world with each region containing multiple data centers in a cluster called Availability Zones. The data centers in these availability zones are interconnected with 100 and 400 gigabit ethernet using AWS designed high density fiber, and redundant providing failover of services within the zone. Utilizing AWS initially gave Riot the ability to deploy more servers at scale locally to players, cutting down on overall latency. AWS has global reach but they don’t have full-fledged regions everywhere, what then?
The next big, next-level thing Riot did is leverage AWS Outposts. AWS Outposts has only been around since 2019 and is still relatively new, but Riot pushed the technology in an entirely new direction. AWS installs an Outpost into a location that you lease out, they connect it to your Wide Area Network (ISP), they manage it, and then it looks just like AWS region in the AWS Console. Riot coordinated these installations with AWS strategically to allow them to put more server resources closer to players (just like in the regions we mentioned earlier). Leveraging AWS Outposts resulted in getting 81% of all players in NA under 35ms and 76% in the EU based on latency to their servers. So now the problem of AWS not having full-fledged regions everywhere slowly becomes mitigated, but what about when they can’t put these outposts everywhere?
While AWS scales much more quickly, Riot is still building out its metal, networking and automation options. While this is a slower process, it’s still important and it’s not just building newer data centers, but iterating against and optimizing what they have. It’s not the preferred option, but Riot has a network of data centers in existence all over the world. The combination of all three options begins to solve the scale and latency problems, but more is still needed to move from the League of Legends model to the lower latency Valorant model.

 

Riot has also continued to optimize Riot Direct which is essentially their own internet backbone as shown in the image above. While Riot Direct has been written about elsewhere, in brief, by strategically purchasing dark fiber and low latency connections to interconnect their infrastructure and peering with multiple upstream providers at different routing points all over the world–and often the same provider at multiple entry points–Riot has direct connectivity to major ISPs all over the world. Having established these peers, Riot Direct uses anycast which advertises the IP address of a game server not just out one of those upstream peers but all of them–this allows that traffic from the player to get routed to the Riot Direct backbone faster where the traffic can be routed internally to the game server location, avoiding the public internet. To get an idea of the scale of what Riot has accomplished, take a look at this link from PeeringDB to see all the different points of entry onto the Riot Direct network all over the world. This network build out over time has resulted in Riot Direct being one of the fastest networks in North America and one of the most connected in the world…not to mention the redundancy to all the locations shown above in the map. Riot Direct also set to add more locations in the Middle East, India and eventually in Africa. Riot effectively has created their own edge network for distributing their content but they didn’t stop there, they decided to integrate AWS and their edge technology into their strategy.

AWS has technology called AWS Global Accelerator (AGA) which is a worldwide solution for handling traffic. When the internet is slow or latency is high, (AGA) routes traffic optimally to your application (or in our case game server running on AWS) in the most optimal automatically. How does it do this? Well, in part, it leverages the same anycast methodology that Riot Direct is using–providing a singular IP (or letting you use your own) for your application (or game server) but advertising it out multiple peering points for entry on to the AWS network so you get to the closest resource. If latency is detected through one network path, it uses another. If server resources are high on one server, it routes it to another.

 

So how good is Riot Direct? According to Ashwin Raghuraman/Solutions Architect at AWS “Riot Direct actually has some regions where it’s better than Amazon Global Accelerator”. To complete the picture of how to create the best experience for Valorant players–and position their network to handle future games with the same requirements–they decided create a hybrid construct that could utilize the best of their network and the global reach of AGA so the two could overlap and complement each other.

AGA and Riot Direct have both have strengths in certain parts of the world and so Riot decided to utilize both architectures simultaneously. Valorant players would send latency tests to both edge networks. Whichever edge network is best for each player, they connect to that network. That is a new enhancement as of November 2021 that gives the individual player (not the group) the ability to be routed to the best location for them personally.

A couple of big take aways from this. The Riot Direct network, combined with AWS Global Accelerator and the build of out of AWS Outposts resulted in significant latency reduction. The EU region that had 76% of players with latency below 35 ms jumped to 85% But this isn’t just for Valorant–this network architecture is already being used for other Riot Games…and Project L is one of them. The gameplay experience that fighting game players will have when they play Project L is going to be like no other fighting game before it. The amazing thing is that not only is Riot continuing to iterate and optimize the network (as their enhancements in November showed) but also as AWS improves and expands AGA and Riot deploys more Outposts the gameplay experience becomes more and more native.

I couldn’t cover everything here (I just hit the highlights here really), but hopefully this high level look into what Riot did Valorant provides insights into why Project L will have the most advanced network capabilities of any fighting in history.


Jermaine Tyson

View all posts