How to measure runtime performance overhead of gramine-sgx?

[Icegrave0391]

Hey, I'm using gramine-sgx enabled nodejs and I'd like to compare its performance against the native performance.

My command is:

gramine-sgx nodejs my_worklaod.js

However, it seems gramine-sgx will show the following message, indicating the overhead of TOML parsing is included:

Gramine is starting. Parsing TOML manifest file, this may take some time...

I tried to add sgx.enable_stat = true, but it seems this will only support perf.

In this case, how can I compare the workload execution (i.e., nodejs) runtime performance against the native performance?

[mkow]

One way to do that would be to add a print with current time after the part you want to skip and then parse the output.

[Icegrave0391]

Thank you. Do you mean instrumenting the source code of gramine-sgx and re-compile it?

I currently use the docker environment for gramine, so I’m just wondering if there is any non-intrusive way to do this?

[mkow]

If by "instrumenting" you mean just manually changing, then yes ;)

if there is any non-intrusive way to do this?

Not really, what you want is to add timing measurements in arbitrarily chosen places, so you have to either modify the code or use more complicated dynamic techniques like dynamic instrumentation or scripting a debugger, but both will significantly change performance, so they won't be a good solution for your problem.

[dimakuv]

In this case, how can I compare the workload execution (i.e., nodejs) runtime performance against the native performance?

Can you add the time-measurement code in your Node.js application itself? This will give you the runtime measurement for both native and Gramine-SGX cases, without any need to modify Gramine itself.

That's typically how benchmarks work any way... They print their own statistics.