AWS Re:Invent FaaS, my key takeaways
With the first digital AWS Re:Invent a few weeks behind us, I had the chance to catch up with my backlog of sessions. Having a strong focus on everything Serverless and FaaS related, these were my personal headlines.
1ms Billing Granularity for AWS Lambda
AWS Lambda lets you run code without provisioning or managing servers, and you only pay for what you use. The previous pricing model was to round the execution time up to the nearest 100ms. Now, with the new 1ms billing granularity, execution time is round up to the nearest ms. The savings are especially visible for low latency Lambda executions (sub 100ms execution time) which are often hot lambda executions. This new pricing model also works with AWS Lambda Provisioned Concurrency.
But there are a few exceptions. Provisioned Concurrency is still calculated from the time it is enabled on a Lambda function until it is disabled, rounded up to the nearest 5 minutes. Another exception is Lambda@Edge, which is still metered in at a granularity of 50ms.
AWS Lambda support for up to 10GB of memory
AWS Lambda could previously be provisioned with up to 3008MB of memory, which translated to ~ two vCPUs. This is now increased to a maximum of 10240MB (10GB). This increased capacity allows Lambda to be used for memory-intensive operations like media processing and ETL/batch jobs.
The increase in memory capacity also translates to an increase in the number of maximum vCPUs. With six vCPUs at its disposal, AWS Lambda can perform compute-intensive applications like machine learning, modelling, genomics and HPC faster.
These more demanding use cases were previously run on AWS Fargate or EC2 because of these limitations CPU and Memory limitations. While there are still no GPU enabled lambdas this might be a start of something bigger.
Container Image support for AWS Lambda
With container image support for AWS Lambda, AWS lowers the threshold even more for developers who want to experiment with AWS Lambda. This dramatically improves the Lambda code portability since the same container could be used to run on AWS Lambda, AWS Fargate and AWS ECS. AWS provides base images for the Lambda runtimes that are currently supported, which can be extended to serve specific use cases.
Another benefit of container image support is that it allows more flexibility for your Lambda dependencies. The current limit of the deployment package is 50MB zipped and 250MB unzipped. With container images up to 10GB, you can introduce larger dependencies in your image. Amazon also claims that the startup and cold start time should not be affected.
Container images combined with the 10GB of memory support and pay by the millisecond. Opens AWS Lambda up to lots of new use-cases and makes already existing once easier. For instance, think about running TensorFlow and Puppeteer while this was already possible now at least it’s easier and more performant.
Amazon CloudWatch Lambda Insights general availability
This new CloudWatch opt-in feature allows developers to gain deeper insights on crucial AWS Lambda metrics like Memory Usage, Duration and Concurrent Executions and Function Cost. For example, fine tuning your AWS Lambda memory allocation will immediately reflect in the Function Cost when your lambda is invoked.
This new feature, combined with an exciting new update in the AWS Compute Optimizer to support AWS Lambda and tools like AWS Lambda Power Tuning will allow you to optimise your AWS Lambda functions.
AWS Batch goes serverless with AWS Fargate (Spot) support
With the addition of Fargate and Fargate Spot as computing resources in an AWS Batch compute environment, you no longer need EC2 instances or a fleet manager to spin up a fleet of EC2 Spot instances. While the managed AWS Batch already used ECS as a platform to run the Batch jobs, Fargate makes it even easier to configure and execute batch jobs at scale.
AWS announced many exciting features so be sure to check out the list of features and service updates.
Are you interested in finding out what migrating to the cloud and using serverless functions can mean for your organisation? Get in touch with us at Cloudway.be and join our serverless world!
