Clojurians Log v2

Based on my rough mental calculation, if I render one HTML page at a time with one "synchronous" SCI script in one node.js process, I get a mixed I/O + CPU workload. For mixed I/O + CPU workloads, it seems running 3~5 times as many parallel tasks as the number of CPU cores is good enough for saturating CPU cores. If I rent a cloud computer with 8 CPU cores and 16GB RAM, it doesn't make sense to have more than 40 processes for my use case. With only 40 processes, I can easily set node.js max old size to 128MB. I would still have enough RAM left for other things.... With 256MB for each process, I would need to reduce the process count down to 4 processes per core. It seems this multi-process architecture is good enough for running untrusted SCI scripts against mixed I/O + CPU workloads. I can maybe even think about using babashka processes for server-side SCI HTML rendering instead of node.js. For IO workloads, I don't think you need custom computation which SCI is useful for. Thus, I think babashka/node.js processes are currently the best substrates for untrusted SCI scripts. If you are going to saturate a cloud computer's CPU cores with untrusted SCI scripts, then it makes sense to reserve a computer for SCI scripts. If it's reserved for SCI scripts, just give it enough RAM.