Artifact Review Summary: Partial Failure Resilient Memory Management System for (CXL-based) Distributed Shared Memory

Artifact Details

Badges Awarded

Artifact Available Artifact Functional Results Reproduced
Artifacts Available (v1.1) Artifacts Evaluated - Functional (v1.1) Results Reproduced (v1.1)

Description of the Artifact

The artifact, CXL-SHM, is a distributed memory management system designed to address the challenges of memory management in a distributed environment, particularly in the presence of client failures. It is posted on a GitHub repository. It contains a detailed README.md document, which provides configuration instructions and scripts for each plot mentioned in the paper. The cxl-shm library is included in the repository, with its source code located in the /include and /src directories. Additionally, the repository includes test cases for evaluating the cxl-shm library, which can be found in the /script and /test directories.

Environment(s) Used for Testing

  • Two-socket AMD Server (each socket has 64 physical cores and 12GB DRAM, Ubuntu 22.04)
  • Rack server (two Intel 4th Gen Xeon Platinum 8470 processors with 4800 MHz and 512 GB DDR5 DRAM for each socket): Although the Intel CXL type3 design example did not support the FPGA DEV KIT during the artifact evaluation period, one reviewer managed to set up an numactl-based environment to evaluate the artifact.

Step-By-Step Instructions to Exercise the Artifact

The reviewers managed to follow the instructions in the README file provided by the authors to run the artifact.

How The Artifact Supports The Paper

Artifact Available

The artifact is open-source and publicly accessible at https://github.com/madsys-dev/sosp-paper19-ae.

Artifact Functional

After installing the required dependencies, the cxl-shm library can be compiled without any issues and provides a functional API in a simulation environment.

Artifact Reproduced

The main claim is that cxl-shm can provide better performance with the requirement of failure tolerance on cxl distributed shared memory. After executing the scripts on the machines, the results can support this claim.

Additional Notes and Resources

The reviewers encountered an issue with Phoenix. Although it is a known issue (https://github.com/kozyraki/phoenix/issues/3) with Phoenix, the reviewers had to manually search and fix it. It would be beneficial if this issue could be noted in the document, or the authors could provide a version with the issue fixed.