Salt Security Cuts Spark Workload Costs by 40% with Granulate

Download Case Study

40%

Reduced Costs

15%

Reduced Spark Time

35%

Reduced CPU Utilization

With a groundbreaking security solution that prevents and protects customers' API attacks, Salt Security turned to Granulate to optimize its Spark workload – responsible for driving the machine learning behind its API Context Engine (ACE) and reduce its AWS costs with its non-intrusive, hands-free continuous optimization solution.

About

Salt provides a groundbreaking security solution that prevents API attacks, and uses machine learning and AI to automatically and continuously identify and protect its customers’ APIs. Found at the core of countless SaaS, web, mobile, microservices, and IoT applications, APIs are a new and critical avenue for breaches, and Salt rallies an impressive cloud infrastructure and vast volumes of data to ensure its APIs are airtight.

Salt’s API Protection Platform is a patented solution that is especially effective for its use of behavioral protection, which hinges on advanced proprietary machine learning algorithms run on enormous datasets. Spark worker clusters hosted on AWS elastic Kubernetes infrastructure handle the task of continuously analyzing copies of all API traffic from customer environments, staying ahead of the next generation of attacks before they occur.

We were pleasantly surprised when Granulate exceeded its own estimate for performance improvement. With better latency conditions and more efficient CPU utilization, we were able to get more out of our machines and cut Spark job completion time drastically.

Gal Porat, DevOps Manager

The Challenge

Salt’s primary Spark workload – responsible for driving the machine learning behind its API Context Engine (ACE) – was in need of additional performance and stability, as the company won new customers and existing customers expanded their involvement with Salt by connecting more APIs. Run on many c5.xlarge EKS instances and totalling hundreds overlocked cores, Salt’s Spark jobs were taking more hours and more compute resources each day to process jobs, leading the company to seek out optimization initiatives so it could reduce its AWS costs.

But with an engineering team whose time was best spent on product-led growth, and a vital business workload that could not suffer extra downtime, Salt had little incentive to pursue in-house performance optimization.

The Results

After activating Granulate on its EKS environment, the agent began automatically and continuously optimizing resource management at the kernel and runtime levels.

Salt immediately saw compute time drop 15% on the Spark worker jobs, and the average core count reduced from 2.76 to 1.63. The company translated this 40% performance gain into a significant advantage; as it didn’t need to increase instance count as it processed larger and larger volumes of traffic.