Caging Performance Benchmarks

[rennergade]

Issue to track our desired benchmarks for Caging.

All of these items should compare Native Linux vs Cage times for this. Ideally, we could change the cage or Native stack when doing this so be explicit about the cage technology stack and native stack when providing data. For example, list versions of software and whether this is NaCl + RustPOSIX or Wasm + RawPOSIX, etc.

For each test, also do a reasonable breakdown of where the time cost comes from when possible. This need not be extensive and we can iterate on this later.

Benchmark List:

• fork comparison
• exec comparison
• signal overhead
• context/cage switch overhead
• system call/microvisor call overhead
• LAMP stack total memory use comparison
• Python benchmark (computation only)
• Macro benchmarks of real world application stacks. We should have a cadre of these which we can add to
• Pipe benchmark
• Unix Domain Socket benchmark
• Loopback benchmark
• /proc, /dev/urandom, etc. benchmark

[JustinCappos]

I've updated this. Please let me know if this isn't clear enough

[rennergade]

I've updated this. Please let me know if this isn't clear enough

This is great for now. Thank you for adding these!

[rennergade]

Hopefully we can start building this in the coming weeks.

@m-hemmings Can you look into this to figure out what may be a good infrastructure to use to run benchmarks? We've used criterion for benchmarking things inside of Rust, but here we need something that can benchmark a full program possibly?

[rennergade]

As I posted channel, lmbench may be helpful for some of these.

We can compile individually using our fork here: https://github.com/Lind-Project/lmbench

I know this has benchmarks for context switching, as well as pipes/domain sockets/sockets.