For our Tor experiments, we used to preloading openssl encryption functions in Shadow version 1.x. This used to lead to a big performance boost (see Figure 2 in the original Shadow paper).

Since that, modern CPUs enable optimizations that can perform AES operations with very low latency. We are wondering if it still makes sense to maintain the code needed to preload the AES functions going forward.

To help us decide, we should run an experiment in Shadow 2.x both with and without preloading the AES functions, and test how performance is affected.