A Critique of RDMA

By Patrick Geoffray, Ph.D.

August 18, 2006

Do you remember VIA, the Virtual Interface Architecture? I do. In 1998, according to its promoters — Intel, Compaq, and Microsoft — VIA was supposed to change the face of high-performance networking. VIA was a buzzword at the time; Venture Capital was flowing, and startups multiplying. Many HPC pundits were rallying behind this low-level programming interface, which promised scalable, low-overhead, high-throughput communication, initially for HPC and eventually for the data center. The hype was on and doom was spelled for the non-believers.

It turned out that VIA, based on RDMA (Remote Direct Memory Access, or Remote DMA), was not an improvement on existing APIs to support widely used application-software interfaces such as MPI and Sockets. After a while, VIA faded away, overtaken by other developments.

VIA was eventually reborn into the RDMA programming model that is the basis of various InfiniBand Verbs implementations, as well as DAPL (Direct Access Provider Library) and iWARP (Internet Wide Area RDMA Protocol). The pundits have returned, VCs are spending their money, and RDMA is touted as an ideal solution for the efficiency of high-performance networks.

However, the evidence I'll present here shows that the revamped RDMA model is more a problem than a solution. What's more, the objective that RDMA pretends to address of efficient user-level communication between computing nodes is already solved by the two-sided Send/Recv model in products such as Quadrics QsNet, Cray SeaStar (implementing Sandia Portals), Qlogic InfiniPath, and Myricom's Myrinet Express (MX).

Send/Recv versus RDMA

The difference between these two paradigms, Send/Receive (Send/Recv) and RDMA, resides essentially in the way to determine the destination buffer of a communication.

  • In the Send/Recv model, the source issues a Send that describes the location of the data to be sent, the destination posts a Receive that similarly indicates where the data is going to be written, and a matching capability is used to associate a posted Recv to an incoming Send. Each side has part of the information required for the completion of the communication; this is a “two-sided” interface.
  • In the RDMA model, both origin and destination buffers must be registered prior to any operations. This memory registration returns a handle that can be used in RDMA operations, Read or Write (a.k.a. Get or Put), to describe the origin or destination buffer. Only one side needs to have all of the information required for the completion of the communication; this is a “one-sided” interface.

In fact, most high-speed interconnects support both Send/Recv and RDMA (Put/Get) programming models, but the Send/Recv implementations of RDMA-based fabrics often lack a rich matching capability, which is a key element.


MPI is a two-sided interface with a large matching space: a MPI Recv is associated with a MPI Send according to several criteria such as Sender, Tag, and Context, with the first two possibly ignored (wildcard). Matching is not necessarily in order and, worse, a MPI Send can be posted before the matching MPI Recv is ready, creating the need to handle unexpected messages.

It is trivial to layer a Send/Recv interface on top of another Send/Recv interface as long as its matching space is wide enough. MPI requires 64 bits of matching information, and MX, Portals, and QsNet provide such a matching capability.

InfiniBand Verbs and other RDMA-based APIs do not support matching at all: they associate incoming Sends with outstanding Recvs in the order in which the Recvs were posted. If only one process can send to a specific receive queue, then you need to send messages in the same order you want to receive them. In the case of multiple senders (Shared Receive Queue), then you have no guarantee about order between senders. Worse, not having a Recv posted at the time an incoming message arrives is considered a fatal error in reliable mode. This lack of support for unexpected messages requires a specific flow-control mechanism to avoid this situation.

When implementing MPI on top of a RDMA programming model, you need to do the matching explicitly by sending a message containing the matching information to the receive side.

Overhead (zero-copy)

One of the key alleged benefits of RDMA is to avoid intermediate memory copies. How does that work with MPI? Not well, for two reasons: memory registration and synchronization.

Memory registration

Zero-copy requires both origin and destination buffers to be registered to allow safe access by the network device. Memory registration and deregistration are expensive operations. They involve a system call and an iteration over each virtual page in an OS-specific way. This registration overhead is actually similar to and sometimes exceeds the cost of a simple memory copy, going against the original objective of avoiding the overhead of intermediate copies.

So, is it possible to do zero-copy in MPI? Yes, but in very specific situations or with some unpleasant tricks:

  • Using lightweight kernels without virtual memory (Cray XT3, IBM Blue Gene).
  • Patching the various Linux kernels to implement safe long-term registration (Quadrics).
  • Hijacking the Malloc library to implement a non-portable and potentially unsafe user-level registration cache (MPICH-GM, MVAPICH, OpenMPI)
  • Optimize the registration code with specific NIC support so that the overhead is cheaper than the memory copy for large messages (MX).

In short, zero-copy, thus RDMA, is practical and efficient in the common case only when the memory registration can occur out of the critical path, which is incompatible with MPI semantics as well as Sockets semantics.


To do zero-copy, you also need to know where you are going to read/write the data on the remote side. Inasmuch as MPI is a two-sided interface, you need to do the matching prior to the RDMA operation, thus requiring an explicit synchronization of both the send and receive sides. The most dramatic side effect is that the MPI sender has to wait for the MPI receiver to be ready before completing the communication and reusing the send buffer. In some situations, the receive side can be very late and the matching synchronization propagates the delay to the send side. For this reason, it is not worth avoiding an intermediate copy on the send side for small/medium messages. Once the message is buffered, the MPI_Send can complete and the application can effectively move on, including reusing the send buffer.

So, it may be surprising for outsiders that most MPI implementations that claim zero-copy actually do a memory copy on both send and receive sides for small/medium messages, typically up to 16K or 32K messages, to avoid synchronization and memory-registration overhead. The RDMA model does not help in any way. The following graph, generated by my Myricom colleague Dr. Loïc Prylli, illustrates this trade-off in MX between using registration and deregistration, which provides the lowest overhead for large messages, versus memory copies, which provides the lowest overhead for short/medium messages.

Low Latency

Inasmuch as the RDMA operations are faster in native RDMA interconnects than the implementation of Send/Recv over RDMA, it is logical to try to use RDMA for small messages, even with intermediate copies. The protocol is straightforward: the send buffer is copied into a pre-registered internal buffer, RDMA-Written into a pre-registered internal buffer on the receive side, and then copied out to the corresponding MPI receive buffer. The problem lies in the one-sided characteristic of the RDMA Write: only one process can safely write into a remote buffer at a time. So the solution is to allocate a pre-registered internal buffer for each possible sender and look for messages in all of these separate buffers. Sounds nice? Not really. This strategy does not scale in time or in space. With only two processes, the receive side has only one buffer to poll. With N processes in the MPI job, the receive side has (N-1) locations to poll. Inasmuch as the polling time increases linearly with the number of processes, so does the latency. That's why most MPI implementations on RDMA-based fabrics use an RDMA method for small jobs, and fall back to a slower, more scalable, Send/Recv method for larger MPI jobs. For example, here are the short-message latencies reported in “InfiniBand Scalability in Open MPI“, Shipman, et al, IPDPS, May 2006, for the two modes of MVAPICH over InfiniBand.

What does it mean practically? Point-to-point micro-benchmarks (i.e., marketing benchmarks) are artificially better on small configurations, but performance is different in networks of useful scale. By contrast, all Send/Recv interconnects have consistent latencies, independent of the number of processes in the MPI job. Also, even the 4.19 us latency above compares unfavorably with the short-message latencies of QsNet, Portals/SeaStar, InfiniPath, or MX/Myri-10G.

Overlap of Communication and Computation (Progression)

Inasmuch as MPI is a two-sided interface requiring matching, it is important that this matching can occur independently of the MPI application, particularly for large messages. Without such asynchronous progress, it is not possible to overlap communication with computation, even when doing zero-copy.

The two following graphs are reproduced by permission from “Measuring MPI Send and Receive Overhead and Application Availability in High Performance Network Interfaces,” Doerfler, et al, EuroPVM/MPI, September 2006, a paper I highly recommend. Rather than showing overhead, the graphs show the send and receive overlap capability — the fraction of the processor time available for computation — with several recent HPC systems, listed here with their interconnects:

  • Red Storm: Cray SeaStar/Portals
  • Tbird: Cisco InfiniBand/MVAPICH
  • CBC-B: Qlogic InfiniPath
  • Odin: Myricom Myri-10G/MPICH-MX
  • Squall: Quadrics QsNetII

Asynchronous progress can be implemented by matching completely in the NIC (Quadrics QsNet and Cray SeaStar) or in the host with specific NIC support (Myricom MX). Some sort of progression could be done on top of a RDMA programming model, but would have a significant impact on short-message latency. Because matching is an integral part of the Send/Recv model, asynchronous implementation is easier in this context. Note that InfiniPath, even though it uses the Send/Recv model, always does a copy on the receive side for large messages, so it cannot overlap.

The graphs above were produced using the Sandia MPI Micro-Benchmark Suite (SMB).


The biggest can of worms that the RDMA programming model brings to the table relates to scalability. RDMA is a connected model, i.e., one software entity called a Queue Pair (QP) in one process space is associated with one and only one such entity in a different process space. So, if you want to be able to communicate with all N processes in your MPI job, you need to use a different QP for each process except yourself. That's (N-1) QPs per process. Even with multiple processes per nodes, the number of QPs per RDMA-based NIC is large enough to accommodate all of them. However, this connected model directly affects the host memory footprint: each QP consumes host memory, so the total memory footprint grows linearly with the number of QPs. For example, it was not uncommon to see several gigabytes of memory required for large InfiniBand clusters. The following graph, which well illustrates this point, is from “InfiniBand Scalability in Open MPI“, Shipman et al, IPDPS, May 2006.

By contrast, the memory footprint of all Send/Recv-based interconnects is constant and independent of the number of processes in the MPI job.

Recently, the addition of the Shared Receive Queue (SRQ) concept, noted in the “Open MPI — SRQ” in the graph above, alleviates some of this scalability nightmare: a receive queue can be shared among several connections, thus consolidating the posted buffers used for the limited Send/Recv functionality. It is interesting to note that a shared receive queue is inherently a Send/Recv programming model feature, perhaps a sign that the native RDMA interconnects are slowly migrating to the more efficient Send/Recv programming model. However, SRQ does not address the overhead of initially establishing the connections, so a connection-on-demand mechanism is needed to hide the problem. This approach may work with MPI microbenchmarks, in which processes typically do not have to communicate with each peer in the MPI application, but this approach will not work well for many real-world MPI applications.


The RDMA programming model does not fit well with the two-sided semantic constraints of MPI or the Sockets interface. It's nothing new; it has been demonstrated before with previous incarnations of RDMA-based low-level communication layers such as VIA. However, the hype is on and the marketing message is wrongly promoting RDMA as the ultimate unifying programming model. Is the RDMA model completely useless? Of course not. It can be leveraged successfully for one-sided programming paradigms such as ARMCI (Aggregate Remote Memory Copy Interface) or UPC (Universal Parallel C), but these paradigms have yet to show better performance and ease of use than MPI.

History always repeats itself, so we can expect the current RDMA programming model to continue to struggle alongside the efficient Send/Recv model, and eventually to fade away.

The author would like to thank Charles Seitz and Loïc Prylli for their help in writing this article.


Patrick Geoffray earned his Ph.D. at the University of Lyon, France, in 2001. His interests lie in high-performance computing. He is currently a Senior Software Architect at Myricom, in the Software Development Lab in Oak Ridge, TN. He is responsible for the design of the Myrinet Express (MX) message-passing system and its associated firmware implementations, and was earlier in charge of various middleware running on GM/Myrinet, including MPICH-GM and VIA-GM.

Subscribe to HPCwire's Weekly Update!

Be the most informed person in the room! Stay ahead of the tech trends with industy updates delivered to you every week!

At SC19: What Is UrgentHPC and Why Is It Needed?

November 14, 2019

The UrgentHPC workshop, taking place Sunday (Nov. 17) at SC19, is focused on using HPC and real-time data for urgent decision making in response to disasters such as wildfires, flooding, health emergencies, and accidents. We chat with organizer Nick Brown, research fellow at EPCC, University of Edinburgh, to learn more. Read more…

By Tiffany Trader

China’s Tencent Server Design Will Use AMD Rome

November 13, 2019

Tencent, the Chinese cloud giant, said it would use AMD’s newest Epyc processor in its internally-designed server. The design win adds further momentum to AMD’s bid to erode rival Intel Corp.’s dominance of the glo Read more…

By George Leopold

NCSA Industry Conference Recap – Part 1

November 13, 2019

Industry Program Director Brendan McGinty welcomed guests to the annual National Center for Supercomputing Applications (NCSA) Industry Conference, October 8-10, on the University of Illinois campus in Urbana (UIUC). One hundred seventy from 40 organizations attended the invitation-only, two-day event. Read more…

By Elizabeth Leake, STEM-Trek

Cray, Fujitsu Both Bringing Fujitsu A64FX-based Supercomputers to Market in 2020

November 12, 2019

The number of top-tier HPC systems makers has shrunk due to a steady march of M&A activity, but there is increased diversity and choice of processing components with Intel Xeon, AMD Epyc, IBM Power, and Arm server ch Read more…

By Tiffany Trader

Intel AI Summit: New ‘Keem Bay’ Edge VPU, AI Product Roadmap

November 12, 2019

At its AI Summit today in San Francisco, Intel touted a raft of AI training and inference hardware for deployments ranging from cloud to edge and designed to support organizations at various points of their AI journeys. The company revealed its Movidius Myriad Vision Processing Unit (VPU)... Read more…

By Doug Black

AWS Solution Channel

Making High Performance Computing Affordable and Accessible for Small and Medium Businesses with HPC on AWS

High performance computing (HPC) brings a powerful set of tools to a broad range of industries, helping to drive innovation and boost revenue in finance, genomics, oil and gas extraction, and other fields. Read more…

IBM Accelerated Insights

Help HPC Work Smarter and Accelerate Time to Insight


[Attend the IBM LSF & HPC User Group Meeting at SC19 in Denver on November 19]

To recklessly misquote Jane Austen, it is a truth, universally acknowledged, that a company in possession of a highly complex problem must be in want of a massive technical computing cluster. Read more…

SIA Recognizes Robert Dennard with 2019 Noyce Award

November 12, 2019

If you don’t know what Dennard Scaling is, the chances are strong you don’t labor in electronics. Robert Dennard, longtime IBM researcher, inventor of the DRAM and the fellow for whom Dennard Scaling was named, is th Read more…

By John Russell

Cray, Fujitsu Both Bringing Fujitsu A64FX-based Supercomputers to Market in 2020

November 12, 2019

The number of top-tier HPC systems makers has shrunk due to a steady march of M&A activity, but there is increased diversity and choice of processing compon Read more…

By Tiffany Trader

Intel AI Summit: New ‘Keem Bay’ Edge VPU, AI Product Roadmap

November 12, 2019

At its AI Summit today in San Francisco, Intel touted a raft of AI training and inference hardware for deployments ranging from cloud to edge and designed to support organizations at various points of their AI journeys. The company revealed its Movidius Myriad Vision Processing Unit (VPU)... Read more…

By Doug Black

IBM Adds Support for Ion Trap Quantum Technology to Qiskit

November 11, 2019

After years of percolating in the shadow of quantum computing research based on superconducting semiconductors – think IBM, Rigetti, Google, and D-Wave (quant Read more…

By John Russell

Tackling HPC’s Memory and I/O Bottlenecks with On-Node, Non-Volatile RAM

November 8, 2019

On-node, non-volatile memory (NVRAM) is a game-changing technology that can remove many I/O and memory bottlenecks and provide a key enabler for exascale. That’s the conclusion drawn by the scientists and researchers of Europe’s NEXTGenIO project, an initiative funded by the European Commission’s Horizon 2020 program to explore this new... Read more…

By Jan Rowell

MLPerf Releases First Inference Benchmark Results; Nvidia Touts its Showing

November 6, 2019

MLPerf.org, the young AI-benchmarking consortium, today issued the first round of results for its inference test suite. Among organizations with submissions wer Read more…

By John Russell

Azure Cloud First with AMD Epyc Rome Processors

November 6, 2019

At Ignite 2019 this week, Microsoft's Azure cloud team and AMD announced an expansion of their partnership that began in 2017 when Azure debuted Epyc-backed instances for storage workloads. The fourth-generation Azure D-series and E-series virtual machines previewed at the Rome launch in August are now generally available. Read more…

By Tiffany Trader

Nvidia Launches Credit Card-Sized 21 TOPS Jetson System for Edge Devices

November 6, 2019

Nvidia has launched a new addition to its Jetson product line: a credit card-sized (70x45mm) form factor delivering up to 21 trillion operations/second (TOPS) o Read more…

By Doug Black

In Memoriam: Steve Tuecke, Globus Co-founder

November 4, 2019

HPCwire is deeply saddened to report that Steve Tuecke, longtime scientist at Argonne National Lab and University of Chicago, has passed away at age 52. Tuecke Read more…

By Tiffany Trader

Supercomputer-Powered AI Tackles a Key Fusion Energy Challenge

August 7, 2019

Fusion energy is the Holy Grail of the energy world: low-radioactivity, low-waste, zero-carbon, high-output nuclear power that can run on hydrogen or lithium. T Read more…

By Oliver Peckham

Using AI to Solve One of the Most Prevailing Problems in CFD

October 17, 2019

How can artificial intelligence (AI) and high-performance computing (HPC) solve mesh generation, one of the most commonly referenced problems in computational engineering? A new study has set out to answer this question and create an industry-first AI-mesh application... Read more…

By James Sharpe

Cray Wins NNSA-Livermore ‘El Capitan’ Exascale Contract

August 13, 2019

Cray has won the bid to build the first exascale supercomputer for the National Nuclear Security Administration (NNSA) and Lawrence Livermore National Laborator Read more…

By Tiffany Trader

DARPA Looks to Propel Parallelism

September 4, 2019

As Moore’s law runs out of steam, new programming approaches are being pursued with the goal of greater hardware performance with less coding. The Defense Advanced Projects Research Agency is launching a new programming effort aimed at leveraging the benefits of massive distributed parallelism with less sweat. Read more…

By George Leopold

AMD Launches Epyc Rome, First 7nm CPU

August 8, 2019

From a gala event at the Palace of Fine Arts in San Francisco yesterday (Aug. 7), AMD launched its second-generation Epyc Rome x86 chips, based on its 7nm proce Read more…

By Tiffany Trader

D-Wave’s Path to 5000 Qubits; Google’s Quantum Supremacy Claim

September 24, 2019

On the heels of IBM’s quantum news last week come two more quantum items. D-Wave Systems today announced the name of its forthcoming 5000-qubit system, Advantage (yes the name choice isn’t serendipity), at its user conference being held this week in Newport, RI. Read more…

By John Russell

Ayar Labs to Demo Photonics Chiplet in FPGA Package at Hot Chips

August 19, 2019

Silicon startup Ayar Labs continues to gain momentum with its DARPA-backed optical chiplet technology that puts advanced electronics and optics on the same chip Read more…

By Tiffany Trader

Crystal Ball Gazing: IBM’s Vision for the Future of Computing

October 14, 2019

Dario Gil, IBM’s relatively new director of research, painted a intriguing portrait of the future of computing along with a rough idea of how IBM thinks we’ Read more…

By John Russell

Leading Solution Providers

ISC 2019 Virtual Booth Video Tour


Intel Confirms Retreat on Omni-Path

August 1, 2019

Intel Corp.’s plans to make a big splash in the network fabric market for linking HPC and other workloads has apparently belly-flopped. The chipmaker confirmed to us the outlines of an earlier report by the website CRN that it has jettisoned plans for a second-generation version of its Omni-Path interconnect... Read more…

By Staff report

Kubernetes, Containers and HPC

September 19, 2019

Software containers and Kubernetes are important tools for building, deploying, running and managing modern enterprise applications at scale and delivering enterprise software faster and more reliably to the end user — while using resources more efficiently and reducing costs. Read more…

By Daniel Gruber, Burak Yenier and Wolfgang Gentzsch, UberCloud

Dell Ramps Up HPC Testing of AMD Rome Processors

October 21, 2019

Dell Technologies is wading deeper into the AMD-based systems market with a growing evaluation program for the latest Epyc (Rome) microprocessors from AMD. In a Read more…

By John Russell

Rise of NIH’s Biowulf Mirrors the Rise of Computational Biology

July 29, 2019

The story of NIH’s supercomputer Biowulf is fascinating, important, and in many ways representative of the transformation of life sciences and biomedical res Read more…

By John Russell

Xilinx vs. Intel: FPGA Market Leaders Launch Server Accelerator Cards

August 6, 2019

The two FPGA market leaders, Intel and Xilinx, both announced new accelerator cards this week designed to handle specialized, compute-intensive workloads and un Read more…

By Doug Black

When Dense Matrix Representations Beat Sparse

September 9, 2019

In our world filled with unintended consequences, it turns out that saving memory space to help deal with GPU limitations, knowing it introduces performance pen Read more…

By James Reinders

With the Help of HPC, Astronomers Prepare to Deflect a Real Asteroid

September 26, 2019

For years, NASA has been running simulations of asteroid impacts to understand the risks (and likelihoods) of asteroids colliding with Earth. Now, NASA and the European Space Agency (ESA) are preparing for the next, crucial step in planetary defense against asteroid impacts: physically deflecting a real asteroid. Read more…

By Oliver Peckham

Cerebras to Supply DOE with Wafer-Scale AI Supercomputing Technology

September 17, 2019

Cerebras Systems, which debuted its wafer-scale AI silicon at Hot Chips last month, has entered into a multi-year partnership with Argonne National Laboratory and Lawrence Livermore National Laboratory as part of a larger collaboration with the U.S. Department of Energy... Read more…

By Tiffany Trader

  • arrow
  • Click Here for More Headlines
  • arrow
Do NOT follow this link or you will be banned from the site!
Share This