Remus is a collection of useful utilities for research-oriented programming. This is the ethos of this project, to accumulate helpful tools that can serve as the basis for development.

Remus Supports:

• Workload driver library (see remus/workload) for experimental evaluation
• Logging utilities (see remus/logging)
• Measurements library (see remus/metrics)
• Sundry other utilities that we don't know where to place yet (see remus/util)
• RDMA (see remus/rdma)
• NUMA (see remus/numa)

Building

We have tested the following configurations:

To build/run your machine requires:

• protobuf-compiler
• librdmacm-dev
• ibverbs-utils
• libspdlog-dev
• libfmt-dev
• nlohmann-json3-dev
• libnuma-dev (if compiling NUMA support)
• CUDA 12.3 (if compiling GPU support)
• doxygen (for building DOCS)
• cmake (3.18 or later)

Your GPU must be Volta or later.

We have the following configuration options/flags:

• GPU (ON or OFF) will compile with GPU support
• KEEP (ON or OFF) will keep ptx and cubin files
• DOCS (ON or OFF) will create documentation
• NUMA (ON or OFF) will enable remus::numa
• LOG_LEVEL (TRACE, DEBUG, INFO, WARN, ERROR, CRITICAL, or OFF) for logging
• CXX_STANDARD (20 or 23) for the C++ standard
• CUDA_ARCHITECTURES (semicolon seperated list of SM numbers)

Using Remus

1. Launch experiment on RDMA-capable Cloudlab cluster
   • Deploy on Ubuntu 24.04 for native gcc-13 support.
2. Edit cloudlab_common.sh
   • Fill out machines, domain, user, and your cloudlab sshkey path.
3. bash cloudlab_install_deps.sh
   • This will install the dependencies on the remote machines
4. bash cloudlab_rebuild.sh
   • Sends over the pre-compiled binary to the remote machines
   • Configuration for the multi-screen setup
   • run screen -c run.screenrc to launch the experiment
   • run screen -c dev.screenrc to log in to all machines

build.sh can also be used to compile the binary locally. Here, you can configure the log level (DEBUG or RELEASE).