This course aims to teach learners about re-envisioning remote memory access for multi-tenant datacenters. The learning outcomes include understanding the challenges of RDMA, exploring the 1RMA approach, and grasping key ideas such as connection-free independent operations and first-class support for security operations. The course focuses on developing skills in addressing connection scalability, security, congestion control, and loss recovery issues. The teaching method involves presenting theoretical concepts and practical solutions. This course is intended for individuals interested in datacenter technologies, network optimization, and remote memory access solutions.
SIGCOMM 2020: 1RMA: Re-envisioning Remote Memory Access for Multi-Tenant Datacenters
Association for Computing Machinery (ACM) via YouTube
Overview
Syllabus
1RMA: Re-envisioning Remote Memory Access for Multi-tenant Datacenters
Problem Statement RDMA is an attractive option for modern datacenter applications due to its low latency and high throughput benefits
Pathologies due to Connections Address connection scalability and performance issues via connection sharing - Multiplexing independent workloads on the same connection
Connection-centric Security
Rigid Congestion Control and Loss Recovery
RDMA Challenges: Root Causes
1RMA Approach Judicious division of labor between hardware and software leading to a simple and fixed-function 1RMA NIC aided by 1RMA software
1RMA Key Idea #1: Connection-Free Independent Ops 1RMA NIC acts on fixed-sized ops and treats them independently
1RMA Key Idea #2: Explicitly-Finite NIC Resource Pools
1RMA Key Idea #3: Solicitation
1RMA Key Idea #5: First-class Support for Security Ops 1RMA offers a Rekey op that provides support for encryption key rotation
What role does local congestion reaction play?
Summary
Taught by
Association for Computing Machinery (ACM)
