RDM-01 · RDMA and HPC Interconnect
Move memory across the network without asking the CPU.
Write verbs programs for send and receive, one-sided reads and writes, and atomics, then tune them until the numbers move.
You build: a verbs client and server with a benchmark and variance figures.
Prefer to be taught it live? RDMA & AI Cluster Networking (AIC-210, 4 days) is the same material delivered in a scheduled cohort by an instructor, in person or online. See the live course · Live dates
In final review — the chapters for this course are still being finalised.
Promise
Write verbs programs for two-sided, one-sided, and atomic operations, and tune them.
Prerequisites
C; basic networking.
Units
1. Overview and ecosystem ch 01–03
what RDMA is and the hardware and software around it.
2. Environment ch 04–05
setting up a development machine and the core resources.
3. The verbs model ch 06–08
the API, memory registration, DMA.
4. Operations and connections ch 09–11
operation types, connection management, hello world.
5. Operation patterns ch 12–15
RDMA write, two-sided operations, completions, atomics.
6. Scaling ch 16–18
multiple QPs, shared receive queues, memory windows.
7. Reliability and tuning ch 19–22
flow control, performance tuning, RoCEv2 configuration.
8. Applications ch 23–26
multi-tenancy, storage, AI and HPC use, troubleshooting.
9. Advanced and testing ch 27–30
advanced models, benchmarking.
Labs
Each lab names its environment and a pass check. Labs are not marked complete on "it ran" — the pass check has to hold.
| Lab | What you do | Environment |
|---|---|---|
| RDM-01ach 11 | Hello-world RDMA on SoftRoCEPass check: Send completes and the message arrives | sim60 min |
| RDM-01bch 12 | RDMA write with a remote keyPass check: Remote buffer holds the written data | sim60 min |
| RDM-01cch 15 | Atomic fetch-and-add on a shared counterPass check: Final value equals the number of clients | sim60 min |
| RDM-01dch 29 | Latency and bandwidth testPass check: Results table with two runs | hardware90 min |
Capstone project · CAP-03
MPI Application Benchmark and Tuning · 3 weeks
A research team's MPI job runs slower on the RoCEv2 fabric than on the old InfiniBand cluster.
Benchmark the application's communication, find the dominant pattern, tune transport and collective choices, and report the gain with its uncertainty.
Deliverables: Benchmark scripts, before-and-after table, a recommendation with risks.
Outcomes
- I can register memory and post a send, a receive, and an RDMA write.
- I can handle completion events correctly without leaking work requests.
- I can use atomics for a shared counter.
- I can measure and explain a latency result.
Also in
How it runs
Work through the units in order. Each unit groups the chapters of the study guide with its share of the labs. You can start any time; there is no cohort date and nothing expires.
Labs run in a VM, a container, a simulator or on real hardware, and the card above names which. Where a lab needs hardware — an RDMA NIC, a programmable switch model, an FPGA — there is a simulator fallback so you are never blocked by equipment.
Questions
Do I get a certificate?
You get a completion record for the course. Path badges and project credentials are earned by passing the labs and the project review, not by attendance.
Can my team take this together?
Yes — and if you would rather have it delivered live, with the instructor in the room, this course also exists as a scheduled live cohort. See the live dates.
How long does it take?
≈ 30 h of study including the labs. Most engineers take it part-time over a few weeks.