rmontroy
commented
4 years ago Here's a reference to how "AWS Lambda appears to treat instance placement as a bin-packing problem, and tries to place a new function instance on an existing active VM to maximize VM memory utilization rates." Lambda uses Linux cgroups to limit CPU, memory, disk I/O, and network, so the memory size parameter actually determines the worst-case allocation of all of these resources. This makes it difficult to get a feel for typical performance of Lambda functions without extensive benchmarking.
I'm pretty sure that the tile cache improves tile load latency much more than maxing out the Lambda memory setting, since the former doesn't involve CPU-intensive JPEG decoding/encoding, but it'd be nice to quantify the difference. It'd also be nice to know the difference in service costs.