Artifacts

Please acknowledge properly in your paper, and kindly drop me an email, if you made use of the following artifacts in your work.

Software

• Echo is our simulator for distributed training that supports 3D parallelism and ex-situ workload tracing. It works for Megatron, PyTorch, and DeepSpeed.

• Code for GOMathL2O in CVPR’24, our gradient-only learning to optimize (L2O) method. Also contains implementations of many baseline L2O methods.

• Testbed and simulation code for Flash, an offchain routing protocol we developed, and traces of two offchain networks (topologies and transactions data) we collected

• Elasecutor, an elastic executor scheduler for data analytics systems based on Spark.

• RepNet, an implementation of flow replication based on node.js.

• NS-3 simulation code for RepFlow on TCP only, based on the ns-3-dcn branch.

Datasets

• Contracts of utilities powering Google's U.S. datacenters I collected in June 2013.

• Dataset of Internet eXchange Points (IXP) traffic I collected in 2012.