Optimizing Serverless Architecture Using Lambda Provisioned Concurrency on AWS with Smartsheet

Smartsheet provides project management tools for nonprofit organizations, small businesses, and over 80 percent of Fortune 500 companies. Built on an event-driven architecture, it processes user activity in near real time (NRT) to facilitate concurrent document editing. To support such NRT collaboration, Smartsheet must minimize latency to promote data consistency and a responsive user experience.

Within minutes, workloads can surge from hundreds to tens of thousands of events per second. To accommodate such fluctuations in traffic, Smartsheet needs a scalable architecture. The company uses a serverless event-processing architecture that is powered by AWS Lambda, a service for running code without thinking about servers or clusters. Smartsheet also supports its architecture using Amazon Simple Queue Service (Amazon SQS), which offers fully managed message queuing for microservices, distributed systems, and serverless applications.

AWS Lambda dynamically initializes new runtime environments in response to incoming traffic volumes. During an initialization, Smartsheet’s function code sends requests to external services and database systems, that can introduce a brief period of latency, reflecting the window in which a new runtime environment is set up. The initialization of a new environment is called a cold start. Although cold starts typically affect less than 1 percent of requests, Smartsheet prioritizes its user experience and has strict latency requirements for its architecture. “We wanted to reduce customer-request latency while keeping costs low,” says Abhishek Gurunathan, Senior Director of Engineering at Smartsheet.

To reduce latency and maintain an excellent customer experience, Smartsheet implemented provisioned concurrency in its architecture. Using this feature, developers can set a number of pre-initialized runtime environments—or warm environments—which stand ready to handle requests. Smartsheet uses provisioned concurrency to run initialization code in advance, avoiding latency increase when creating runtime environments on demand.

Provisioned concurrency facilitates architecture that is highly resistant to traffic spikes. When traffic reaches the point of using up the prewarmed runtime environments, additional requests aren’t throttled. Instead, they’re handled using on-demand concurrency—the capability of AWS Lambda to automatically scale to handle incoming requests. That helps Smartsheet maintain high service availability, even during periods of high demand. When traffic surges, AWS Lambda scales up the number of runtime environments required to handle the increased load.

There are two common approaches to implementing provisioned concurrency: setting a static value or using automatic scaling. With a static configuration, developers specify the number of pre-initialized—or warm—runtime environments. That approach, which Smartsheet adopted initially, is effective for architectures that handle predictable traffic patterns.

Although the static configuration worked well during peak-usage periods, it led to overprovisioning during low-usage ones. To address this, Smartsheet adopted the second approach: automatic scaling. This method automatically adjusts the number of prewarmed runtime environments according to actual traffic, improving solution cost efficiency through a dynamic and responsive system of resource allocation.

Provisioned concurrency with automatic scaling was an ideal match for Smartsheet’s fluctuating traffic patterns. At the same time, the company used AWS Graviton processors, custom-designed server processors developed by AWS to provide the best price performance for cloud workloads. Thus, Smartsheet configured its AWS Lambda functions to use an arm64 architecture that is powered by AWS Graviton.

As a result, Smartsheet decreased the frequency of cold starts and optimized performance. “Using AWS Lambda provisioned concurrency with automatic scaling alongside AWS Graviton, we’ve reduced latency by 83 percent and improved cost-efficiency,” says Gurunathan.

With the elasticity and scalability of its solution, Smartsheet can scale its software-as-a-service product according to traffic volume, smoothly handling traffic surges while automatically scaling down in off-peak hours. Bringing provisioned concurrency with automatic scaling to its architecture, the company also reduced latency. Moreover, by switching to an AWS Graviton–based architecture, Smartsheet is saving 20 percent on the gigabyte-second cost of AWS Lambda functions.

By using serverless services such as AWS Lambda and Amazon SQS, the company can focus less on infrastructure management and scaling concerns and more on what matters—delivering business value and continued innovation.