perf-test-initial-commit

[iniesta-08]

What is in this PR?

We have done several runs to evaluate performace of enclaver/nitro-enclaves. As part of these tests I have created a simple GO server that works using a self signed SSL certificate and does a naive crypto operation.

API used to test - https://localhost:8082/hello

Instance Setup

1. Provision a new nitro instance and follow the steps here to setup up nitro_allocator.yaml file https://edgebit.io/enclaver/docs/0.x/guide-first/

2. Modify the file /etc/nitro_enclaves/allocator.yaml to cpu_count: 4

3. Restart the nitro-allocator service using sudo systemctl restart nitro-enclaves-allocator.service

JMeter setup

SSH into your machine and run the following commands

mkdir jmeter
cd jmeter
wget https://archive.apache.org/dist/jmeter/binaries/apache-jmeter-5.5.tgz
tar -xf apache-jmeter-5.5.tgz
sudo amazon-linux-extras install java-openjdk11

Running the tests

The script ./run_in_enclave.sh contains the commands to start the nitro enclave.

Comment out the 4vcpu part of the script if you are using 2cvpu to start enclaves and vice-versa.

1. Start the enclave using the command ./run_in_enclave.sh
2. Run JMeter tests using ./run_jmeter_test.sh

Sample results

With 4vCPUs -
summary = 12000 in 00:00:42 = 283.7/s Avg: 155 Min: 12 Max: 2707 Err: 0 (0.00%)

With 2vCPUs - summary = 12000 in 00:00:41 = 290.3/s Avg: 108 Min: 14 Max: 1755 Err: 0 (0.00%)

As you can see there is no significant difference in TPS by increasing the number of vCPUs.

[eyakubovich]

Thank you for designing and running the first perf test! To give us more context, could you also:

1. include what VM instance type you used
2. have a "control" for the experiment: what do the numbers look like for that VM when this test container is ran outside of the Enclaver in plain Docker.

Thank you!

[robszumski]

Thanks for putting this together, really cool write up.

[iniesta-08]

@eyakubovich Sure, that makes sense; I will provide the docker benchmarks here.

[eyakubovich]

LGTM. Thank you you for your contribution.