Experiment: does BIO_f_buffer help performance with read/write syscalls?
dgoulet has found that openssl generates lots of small read and write syscalls. Maybe we'd do better to have openssl buffer stuff?
BIO_f_buffer can supposedly help us here. We could buffer reads, writes, or both.
We should see whether this helps performance, whether it causes weird bugs, and so on. We should also try to find out how much memory openssl allocates for this stuff, and take that into account for our OOM calculations.