I attended the Serverless Days Sydney 2019 conference. Here are my top take aways.
Cold Starts Are Still An Issue (But Improving)
A cold start is when a cloud function is invoked for the first time or after a long time of no invocations. Cold starts can be significant (up to 10 seconds on some cloud providers). Several speakers stated that cold start times had reduced in 2019, however there were no statistics to back that up unfortunately.
Cold Start Delay Varies Within Cloud Providers
Geshan Manandhar from the ICONIC displayed some interesting graphs showing that Google Cloud Run cold starts were up to 3 seconds whereas Google Cloud Functions could have cold starts up to 10 seconds.
Cloud Functions Calling Cloud Functions Can Lead to Delays
If one cloud function takes 5 seconds to run after a cold start, and that function calls another cold cloud function that takes 5 seconds to run, and that function calls another cold cloud function…you can see how this will add up.
Cost Reduction Can Be Massive
Denis Bauer, Head cloud computing bioinformatics at CSIRO, managed to reduce cloud costs from $4000 per month to $7 per month using serverless cloud functions over traditional methods.
Cloud Functions Can Help Massively with Scale
Ben Kehoe from iRobot talked about 100x scaling. On Christmas Day, lots of people receive iRobots as gifts in a 4 hour window. This places a 100x demand on their systems. Using AWS Lambdas and 30 other cloud services, they handle this increased volume with little human supervision or input.
Speed Increases Can Be Significant
Running concurrent cloud functions is easier that configuring your own virtual machine instances. Denis Bauer reported that time querying 100,000 genomes for 85 million variants was completed in 1 second using cloud functions, compared to the current average of 40 seconds using traditional methods.