iRoute: Local Routing Table-based Workflow Management in Serverless Computing

iRoute is a workflow orchestration and routing system designed to achieve universal 1-hop transfers while maintaining high resource efficiency. To address the limitations of centralized orchestration, iRoute offloads both orchestration and routing capabilities from global to local. It adopts a dual-layer architecture:

• Local routing controllers make routing decisions and manage direct communication between function instances.
• Centralized coordinator ensures global consistency by synchronizing multiple local routing tables.

iRoute runs as a set of OpenFaaS templates and supports IPC, Socket, and RDMA communication.

---

Installation

iRoute requires the following components to be installed:

• OpenFaaS
• ZooKeeper (for global routing table storage)
• k8s-rdma-shared-dev-plugin (for RDMA support)
• sn-db: Database services for Social Network benchmark.

---

Usage

Templates Overview

Template	Role
python3-cc	Coordinator: assists in direct connection setup and synchronizes routing tables with ZooKeeper.
python3-func	Function instance: performs routing and communication directly with other functions.
python3-entry	Workflow entry: handles service discovery and sends workflow requests.

Note: python3-func does not support OpenFaaS Gateway invocation. Always invoke workflows through python3-entry.

---

Configuration Steps

1. Coordinator (python3-cc)

• Define workflow structure in workflow.json.
  • Function names must match the ones deployed in OpenFaaS.
• Update handler.py with ZooKeeper authentication credentials for accessing the global routing table.
• Modify the get_lat_dist function in handler.py to measure and report request latency distribution.

2. Function Instance (python3-func)

• Synchronize and use the same workflow.json.

3. Workflow Entry (python3-entry)

• Update handler.py with:
  • Coordinator function name in OpenFaaS.
  • Entry function name.
  • The generate_data function for test data generation.

---

Benchmarks

The benchmarks directory includes three representative applications, each implemented for IPC, Socket, and RDMA backends:

Application	Description	Path
Social Network	A latency-sensitive application from DeathstarBench that creates post embedded with text, media, links and user tags.	benchmarks/sn
FINRA	A financial application that validates trades based on trade and market data.	benchmarks/finra
ExCamera	A video-processing application that encodes and processes chunks in parallel.	benchmarks/excamera

Each benchmark directory provides:

• A README with deployment and execution steps.
• Prebuilt Docker images available on Docker Hub.
• Configurations for OpenFaaS deployment.

For detailed instructions, see Benchmark Documentation.