Benchmarking / Profiling Recommendations

[LeibnizCapital]

Any recommendation for running microbenchmarks in the enclave? I cannot naively use libtest bench or criterion as normal. I'm simply looking for general guidance or examples of good benchmarks.

[jvz]

Benchmarking might be complicated here as many use cases should use constant-time algorithms to avoid side channel information leakage. Are you looking for benchmarks of individual operations? I'd be interested in how that could work here as well.

[LeibnizCapital]

I'm interested in benchmarking critical operations (microbenchmarking), like a unit tests as benchmarks. In untrusted land, the libtest bencher has a harness for this.

[dingelish]

hi @LeibnizCapital ,

we have some benchmark results in our CCS'19 paper: https://dingelish.com/ccs19.pdf . section 7 testing code is at https://github.com/mesalock-linux/rust-sgx-benchmark

[LeibnizCapital]

This is helpful @dingelish Since you've wrote these benchmarks, I'll cut to the chase... Do you have recommendations for large key/value store? I see the hugemem and db-proxy examples and some kv crates in third-party. I'm both afraid of abusing encrypted memory and making too many external calls. I'd like to better understand what the tradeoffs are to select the best approach.