CausalMesh

This repository contains the prototype system for our VLDB 2025 paper: CausalMesh: A Causal Cache for Stateful Serverless Computing

Project Structure

ccmesh/ - Server code
  causal/ - define vector clock (VC), happens-before relationship, and all causal consistency related functions
  redis/ - causal layer on top of redis
  ipc/ - data serialization and transmission
  scheduler/ - a naive scheduler that randomly schedules functions to different compute nodes
  workload/ - microbenchmark and a real-world application (media app) used in evaluation
  # The following 4 folders implement server APIs defined in Table 4 in the paper
  ccmesh/ - causalmesh server logic
  con/ - hydrocache-con server logic
  opt/ - hydrocache-opt server logic
  faas/ - faasTCC server logic

ccmesh-go/ - Client code in golang, clients are compiled to binary and registered in Nightcore (https://github.com/ut-osa/nightcore)
  the following 4 caches implement Client APIs defined in Section 5.1
  pkg/ccmesh - causalmesh client library (Section 6)
  pkg/hzcon - hydrocache-con client library
  pkg/hzopt - hydrocache-opt client library
  pkg/hzfass - FaaSTCC client library


causal-tla/ - TLA+ model
  mesh.tla - model checking code
  prood.md - lammas and invariants explained in plain English

Experiment Setup (w/ Cloudlab)

CausalMesh Server

Prerequisites

Clone both the server code (ccmesh) and the client code (ccmesh-go) in the same directory, fill in the location of the directory, key file location and cloudlab user name in scripts/common.py

Install Python 3 and fabric

pip install fabric

Setup

Use the following profile to create a cluster on Cloudlab

https://www.cloudlab.us/p/a22d79e1ff06310e24e133d03acc1703ec3b09ad

Copy the hostname of the nodes to scripts/common.py.

Run on the host

python scripts/upload.py
python scripts/setup.py

After setup finishes on all server, log in the first machine on cloudlab (10.10.1.1)

# on 10.10.1.1
sudo sed -i 's/bind 127.0.0.1 -::1//g' /etc/redis/redis.conf
sudo service redis-server restart
echo "CONFIG SET protected-mode no" | redis-cli

Run experiments

To run Causalmesh, HydroCache-Con and HydroCache-Opt

change feature flags in Cargo.toml to

default = ["ccmesh", "cloud"]

default = ["con", "cloud"]

default = ["opt", "cloud"]

respectively, then run

python scripts/upload.py

All following commands should be run on the host.

Microbenchmarks

python scripts/run.py --alg ccmesh --exp micro
python scripts/run.py --alg con --exp micro
python scripts/run.py --alg opt --exp micro

Real world application

python scripts/run.py --alg ccmesh --exp media
python scripts/run.py --alg opt --exp media

Various length of workflow

python scripts/run.py --alg ccmesh --exp scale_lambda

Various number of cache server

python scripts/run.py --alg ccmesh --exp scale_server

Visibility

python scripts/run.py --alg ccmesh --exp vis