How Freshworks Autonomously Optimized its AWS Lambda Developer Platform

Summary

• Freshworks runs 30,000 Lambda functions for its developer platform and needed a way to optimize performance so that end users had positive experiences with apps built on the developer platform
• Freshworks deployed Sedai’s Autonomous Optimization capability to autonomously find the optimal way to configure each Lambda function for performance
• The company realized 34% latency reduction without incurring more in cloud costs, boosting performance for end users

Problem

Freshworks has quickly grown into a company with 1,000 employees, five offices, and more than 130,000 clients using its products ranging from customer support and help desk software to recruitment tools and sales CRM software.  Freshworks has made moves to serve developers to extend the capability of its platform, and hosts thousands of developer apps on its developer platform.

As with all Freshworks services, optimal performance is expected.  “At Freshdesk we are obsessed with providing a great experience to our customers .. speed is critical” shared a Site Reliability Engineer at Freshworks.

As the popularity of the developer platform grew, Freshworks was running 30,000 Lambda functions, but did not have a way of optimizing each service for performance.

Solution

Freshworks deployed Sedai’s Autonomous Optimization capability in order to keep each Lambda function optimized.  Sedai was connected to AWS CloudWatch metrics and began to use the data to understand the performance and cost for different configurations.Based on high level direction as to whether to pursue cost, performance or a mix of goals, Sedai is able to both perform initial optimizations and make continued optimizations as new versions of the function are shipped and as application traffic changes.

Results

Improved latency

Freshworks was able to find an average 34% latency reduction across their Lambdas, primarily by optimizing  memory settings.  Individual functions showed latency reductions of up to 95%. As a result, Sedai achieved 15 years of execution savings (5,618 days of execution time) as shown in the dashboard at right.

An alternative objective of cost reduction was also evaluated but not pursued; this would have yielded 27% cost savings with no performance change as performance was the priority.

Operational Productivity

The Freshworks Developer Platform Lambda optimization now operates autonomously and the Freshworks team does not need to spend time managing Sedai.

Release Quality

Each new release on the platform is scored based on the changes in cost, latency and errors from the prior version.  The level of deviations from the prior version are reported on a dashboard inside Sedai (see chart at right).

‍