Fast Raft

For our final project for CPSC 426 (Fall 2024), we implemented FastRaft, a variant of the popular Raft consensus algorithm. Fast Raft optimizes the log replication process to introducing a different replication strategy to reduce the average round trip latency between leaders and followers, enabling higher throughput under heavy loads (and lower network failure conditions). We also containerized the Raft and FastRaft nodes and deployed them onto an EKS cluster using Terraform, conducting various correctness and performance benchmarks.

Team Members

• Bryan Sebaraj, bws26, [email protected]
• Anton Melnychuk, am3785, [email protected]

Building and Running

Building docker images:

• Install docker or Docker Desktop
• Raft Node: cd raft && docker buildx build . --platform linux/amd64 -f Dockerfile -t <raft_image_repo> && docker push <raft_image_repo>
• FastRaft Node: cd fastraft && docker buildx build . --platform linux/amd64 -f Dockerfile.fast -t <fastraft_image_repo> && docker push <fastraft_image_repo>
• Load Tester: cd raft && docker buildx build . --platform linux/amd64 -f Dockerfile.load -t <loadtester_image_repo> && docker push <loadtester_image_repo>

Deploying to EKS

• Install terraform, awscli, and kubectl
• Configure your AWS credentials and kubernetes config (create cluster role bindings, etc).
• cd raft/nodegroup
• Modify main.tf to point to image repositories and have correct AWS URLs
• Run: terraform init && terraform plan && terraform apply
  • Note: Remember to update IAM roles on AWS (relatively straightforward on the console/website)

Testing

• To run test scripts:
  • kubectl get pods to get the pod name for the raftload pod
  • kubectl exec -it <raftload_pod_name> -- sh
  • ./app "ME=${SELF_IP}:5000" "PEERS=${PEERS_IP}" to run the load tester on the raft cluster
    • change ${PEERS_IP} to the addresses of the fast raft nodes if testing fast raft
• To change network characterictics of the cluster:
  • kubectl exec -it <raft_pod_name> -- sh to attach to a raft or fast raft node
  • tc qdisc add dev eth0 root netem delay 100ms to add a 100ms delay to the network
  • tc qdisc add dev eth0 root netem loss 1% to add a random packet loss of 1%
  • tc qdisc del dev eth0 root to remove the delay or packet loss
  • mirror commands to other nodes in cluster as desired

Demonstration

The demonstration video can be found via Google Drive here.

Group Work

• Bryan SebaRaj: Containerized Raft/FastRaft nodes, deployed Raft/FastRaft onto EKS via Terraform, added tests and conducted performance/correctness experiments, debugged FastRaft, wrote the README, and filmed the demonstration video. Contributed equally to write-up.
• Anton Melnychuk: Swapped out network layer for gRPC, bootstrapped Raft, and implemented the FastRaft optimizations. Contributed equally to write-up.