FaaSim is a discrete-event simulator for modeling main-stream serverless systems. It disaggregates the three scheduling dimensions: (1) load balancing, (2) autoscaling, and (3) placement. The algorithms of each dimension are pluggable, enabling the exploration of multiple policy choices in one system.

( This is a prototype that I developed as a proof of concept for my research at ETH. For more details: Hongyu He <[email protected]> )

$ pip3 install -r requirements_dev.txt

$ python3 faasim <rps | test>

The following validations were conducted against the popular serverless platform Knative.

For the following experiments, the hardware specifications are the following:

• Cluster size: 2 nodes (1 master + 1 worker)
• Number of cores per node: 16 -> maximum theoretical throughput is 16 requests per second
• Function execution time: 1 s (50 percentile from Azure function trace)
• Function memory footprint: 170 MiB (50 percentile from Azure function trace)

In the following experiments, the system was not warmed up in order to preserve cold start.

• CDF:

• 50 percentile (p50):

• 99 percentile (p99):