Performance Tales of Serverless
summary
Compares AWS Lambda, Azure Functions and Google Cloud Functions with concrete benchmarks: measures cold starts that can add seconds, shows local processing of 1 million events takes seven hours while serverless ranges from ~22 minutes down to 1 million events in 4 minutes on tuned AWS (limited by Kinesis throughput); Google hits concurrency/instance limits and requires HTTP connection pooling fix to handle spikes. Uses Kinesis, Pub/Sub, memory/CPU tuning and load tool Artillery to build and validate the tests.
AI-written summary · about
chapters (26)
- 0:00Intro
- 0:16Function-as-a-Service in one picture
- 1:09Faas Cloud Offerings
- 1:46Why Serverless?
- 3:05Ideal Scalability: Throughput
- 3:26Ideal Scalability: Latency
- 6:33Visitors during sample 5 hours
- 7:02Dependency Call Duration (ms)
- 7:29Target End-to-End Latency (ms)
- 8:07AWS Lambda Latency (ms)
- 8:36Google Cloud Functions Latency (ms)
- 8:52Azure Functions Latency (ms)
- 9:08Cold Start
- 11:36Requests/sec during sample 5 hours
- 12:39How to improve it
- 14:37Serverless ETL
- 15:25AWS, Azure & GCP Processing 1 Min Points
- 17:06AWS Processing Speed per Reserved RAM
- 19:51Requests/sec around TV show
- 20:33Azure Functions Response Percentiles (ms)
- 21:29AWS Lambda Response Percentiles (ms)
- 22:24Google Cloud Functions: Throttled
- 24:33Preparation of this talk
- 25:26Serverless artillery
- 26:40Is Serverless Good for My Next Project?
- 28:04Perfect Scenarios
related talks
37mPerformance Tales of Serverless
01h 00mServerless - reality or BS - notes from the trenches
35mEfficiency of Serverless Architectures with AWS Solutions
15mAlternative Ways To Think About Serverless
41mGo for serverless functions
01h 02mExploring Serverless Options for .NET in Azure, AWS, and Beyond
58mServerless Under the Hood
20mLost in Serverless: AWS Lambda, Google Cloud Function, Azure Functions...