Firework Games Delivers a Smooth In-Game User Experience and Saves Manpower Costs with AWS

As Spark Era is a highly competitive battle royale game, Firework Games needs to deliver consistently low latencies to its players for a fun and fair gaming experience.

Additionally, using artificial intelligence (AI) to generate unique non-player characters (NPCs), Spark Era aims to deliver a personalized gameplay experience for every player. The NPCs are tailored based on players’ prior activities, such as their in-game gear choices and interactions with other players. As such, Spark Era has to ingest and process large amounts of data for its AI-based features. With these in mind, Firework Games turned to AWS for a cloud infrastructure that can deliver on performance, scalability, and cost-effectiveness.  

On AWS, Firework Games reduced its average latency to 160ms from 300ms connecting users from Korea to AWS servers located in US. This was 46 percent lower than on its previous on-premises servers, while also saving up to 30 percent in costs.

With Amazon Elastic Cloud Compute (Amazon EC2) Auto Scaling and Amazon Elastic Graphics, Firework Games easily scales its compute capacity for player traffic. Since it launched, Spark Era has hosted up to 20,000 players concurrently, with near-zero downtime. Amazon EC2 Auto Scaling has also saved its developers about 40 hours in manual infrastructure maintenance and scaling.

More importantly, Firework Games can deliver a level playing field for players globally with AWS Availability Zones and AWS Regions, keeping latencies within the range of 100-160 ms for global users.

Using Amazon CloudFront, a low-latency content delivery network, players were able to download the game content within 10 minutes. During testing on its previous on-premises servers, this took double the time.

To ingest and process the game data for its machine learning models, Firework Games deployed a combination of Amazon Relational Database for MySQL (Amazon RDS for MySQL), Amazon ElastiCache, Amazon Aurora, and AWS Glue. The company estimates that Spark Era handles 114 TB of data per hour on average.

The company accelerated the game development by 80 percent by deploying AWS Deep Learning AMIs. These provide pre-configured environments for Firework Games, allowing it to move straight to development instead of having to set up a deep learning framework and pipelines, which typically takes up to 3 months.

“The AWS team and AWS Enterprise Support has been very helpful in supporting our development of Spark Era and ensuring the best player experience. They guided us on which AWS Regions meet our needs, and advised us on the Amazon EC2 instances needed for our setup. The AWS team also worked with us on a globally scalable design – from latency reduction through region selection, Amazon CloudFront and AWS Global Accelerator implementation, and Transmission Control Protocol (TCP)-based autoscaling strategy to optimize compute resource usage,” said Moses Ip, chief executive officer, Firework Games. “In summary, AWS helped us make the most of our resources, which is vital for us as a startup.” 

Looking ahead, Firework Games plans to integrate Amazon Polly and Amazon Transcribe into Spark Era, with players interacting with NPCs using their voice instead of having to click on a given list of options, building a more immersive gaming experience.