The first step is to build all subkernels required for our evaluation. On your development environment, change into the ime directory. There you will find a script build.sh, which compiles all required subkernels and creates a .tar archive out of them. The script requires that you have the memclave docker container installed. Running the script should take no longer than a minute:

# dev environment: ./ime

./build.sh

Move the ime-client-library folder into your memclave environment, if it is not already present there. Move the subkernels.tar file, produced in the last step, into the ime-client-library folder. Now you should be able to execute the build.sh script, which compiles the guest side portions of our benchmarks. The build process should take no longer than a minute:

# memclave-environment: ./ime-client-library/

./build.sh

With compilation finished, you should be able to run all benchmarks in your memclave environment, except for one which is described below by using the run.sh script. Simply type

# memclave-environment: ./ime-client-library

./run.sh

into your terminal.

There is one other measurement that needs to be taken using a modified version of the TL, which reports performance stats while loading subkernels in MRAM. We provide a modified boot.sh script to boot up the hypervisor with the modified TL. Shut down the memclave environment and boot it up again using the boot-stats.sh script. Once booted, perform the same setup steps, such as loading the kernel driver. Now change into your ime-client-library directory and execute

# memclave-environment (modified TL): ./ime-client-library

./run.sh subk

which will perform the final memclave benchmark.

To measure baseline performance, some of the same benchmarks need to be build on the execution environment. Move the prim-benchmarks folder of the artifact to the execution environment. Then move the ime-client-library folder into the prim-benchmarks folder on your execution environment. The latter is needed because the MRAM benchmark is written agnostic to whether Memclave or UPMEM is used as the guest-side library. There is no build step for the baseline, builds are handled just before benchmark execution. Use the run.sh script to execute all benchmarks:

./run.sh

Both the memclave and UPMEM benchmarks should have generated a file output.tar. Move those to your development environment, and rename them upmem.tar and memclave.tar respectively. Then run the script plotting/plot_all.sh to generate all figures present in the paper, and to print out table contents into the terminal:

./plot_all.sh memclave.tar upmem.tar