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.

Matching

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.

Synchronization

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).

Scalability

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.

Conclusions

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!

South Africa CHPC: Home Grown Dynasty

October 22, 2018

Before the build up to the final event in the 2018 Student Cluster Competition season (the SC18 competition in Dallas), I want to take a moment to write about one of the great inspirational stories of these competitions. Read more…

By Dan Olds

NSF Launches Quantum Computing Faculty Fellows Program

October 22, 2018

Efforts to expand quantum computing research capacity continue to accelerate. The National Science Foundation today announced a Quantum Computing & Information Science Faculty Fellows (QCIS-FF) program aimed at devel Read more…

By John Russell

Democratization of HPC Part 3: Ninth Graders Tap HPC in the Cloud to Design Flying Boats

October 18, 2018

This is the third in a series of articles demonstrating the growing acceptance of high-performance computing (HPC) in new user communities and application areas. In this article we present UberCloud use case #208 on how Read more…

By Wolfgang Gentzsch and Håkon Bull Hove

HPE Extreme Performance Solutions

One Small Step Toward Mars: One Giant Leap for Supercomputing

Since the days of the Space Race between the U.S. and the former Soviet Union, we have continually sought ways to perform experiments in space. Read more…

IBM Accelerated Insights

Join IBM at SC18 and Learn to Harness the Next Generation of AI-focused Supercomputing

Blurring the lines between HPC and AI

Today’s high performance computers are helping clients gain insights at an unprecedented pace. The intersection of artificial intelligence (AI) and HPC can transform industries while solving some of the world’s toughest challenges. Read more…

Penguin Computing Launches Consultancy for Piecing AI Strategies Together

October 18, 2018

AI stands before the HPC industry as a beacon of great expectations, yet market research repeatedly shows that AI adoption is commonly stuck in the talking phase, on the near side of a difficult chasm to cross. In respon Read more…

By Tiffany Trader

South Africa CHPC: Home Grown Dynasty

October 22, 2018

Before the build up to the final event in the 2018 Student Cluster Competition season (the SC18 competition in Dallas), I want to take a moment to write about o Read more…

By Dan Olds

Penguin Computing Launches Consultancy for Piecing AI Strategies Together

October 18, 2018

AI stands before the HPC industry as a beacon of great expectations, yet market research repeatedly shows that AI adoption is commonly stuck in the talking phas Read more…

By Tiffany Trader

When Water Quality—Not Quantity—Hinders HPC Cooling

October 18, 2018

Attention has been paid to the sheer quantity of water consumed by supercomputers’ cooling towers – and rightly so, as they can require thousands of gallons per minute to cool. But in the background, another factor can emerge, bottlenecking efficiency and raising costs: water quality. Read more…

By Oliver Peckham

Paper Offers ‘Proof’ of Quantum Advantage on Some Problems

October 18, 2018

Is quantum computing worth all the effort being poured into it or should we just wait for classical computing to catch up? An IBM blog today posed those questio Read more…

By John Russell

Dell EMC to Supply U Michigan’s Great Lakes Cluster

October 16, 2018

The University of Michigan (U-M) today announced Dell EMC is the lead vendor for U-M’s $4.8 million Great Lakes HPC cluster scheduled for deployment in first Read more…

By John Russell

Houston to Field Massive, ‘Geophysically Configured’ Cloud Supercomputer

October 11, 2018

Based on some news stories out today, one might get the impression that the next system to crack number one on the Top500 would be an industrial oil and gas mon Read more…

By Tiffany Trader

Nvidia Platform Pushes GPUs into Machine Learning, High Performance Data Analytics

October 10, 2018

GPU leader Nvidia, generally associated with deep learning, autonomous vehicles and other higher-end enterprise and scientific workloads (and gaming, of course) Read more…

By Doug Black

Federal Investment in Exascale – What It Really Means

October 10, 2018

Earlier this month, the EuroHPC JU (Joint Undertaking) reached critical mass, and it seems all EU and affiliated member states, bar the UK (unsurprisingly), have or will sign on. The EuroHPC JU was born from a recognition that individual EU member states, and the EU as a whole, were significantly underinvesting in HPC compared to the US, China and Japan, who all have their own exascale investment and delivery strategies (NSCI, 13th 5 Year Plan, Post-K, etc). Read more…

By Dairsie Latimer

TACC Wins Next NSF-funded Major Supercomputer

July 30, 2018

The Texas Advanced Computing Center (TACC) has won the next NSF-funded big supercomputer beating out rivals including the National Center for Supercomputing Ap Read more…

By John Russell

IBM at Hot Chips: What’s Next for Power

August 23, 2018

With processor, memory and networking technologies all racing to fill in for an ailing Moore’s law, the era of the heterogeneous datacenter is well underway, Read more…

By Tiffany Trader

Requiem for a Phi: Knights Landing Discontinued

July 25, 2018

On Monday, Intel made public its end of life strategy for the Knights Landing "KNL" Phi product set. The announcement makes official what has already been wide Read more…

By Tiffany Trader

CERN Project Sees Orders-of-Magnitude Speedup with AI Approach

August 14, 2018

An award-winning effort at CERN has demonstrated potential to significantly change how the physics based modeling and simulation communities view machine learni Read more…

By Rob Farber

House Passes $1.275B National Quantum Initiative

September 17, 2018

Last Thursday the U.S. House of Representatives passed the National Quantum Initiative Act (NQIA) intended to accelerate quantum computing research and developm Read more…

By John Russell

Summit Supercomputer is Already Making its Mark on Science

September 20, 2018

Summit, now the fastest supercomputer in the world, is quickly making its mark in science – five of the six finalists just announced for the prestigious 2018 Read more…

By John Russell

New Deep Learning Algorithm Solves Rubik’s Cube

July 25, 2018

Solving (and attempting to solve) Rubik’s Cube has delighted millions of puzzle lovers since 1974 when the cube was invented by Hungarian sculptor and archite Read more…

By John Russell

D-Wave Breaks New Ground in Quantum Simulation

July 16, 2018

Last Friday D-Wave scientists and colleagues published work in Science which they say represents the first fulfillment of Richard Feynman’s 1982 notion that Read more…

By John Russell

Leading Solution Providers

HPC on Wall Street 2018 Booth Video Tours Playlist

Arista

Dell EMC

IBM

Intel

RStor

VMWare

TACC’s ‘Frontera’ Supercomputer Expands Horizon for Extreme-Scale Science

August 29, 2018

The National Science Foundation and the Texas Advanced Computing Center announced today that a new system, called Frontera, will overtake Stampede 2 as the fast Read more…

By Tiffany Trader

HPE No. 1, IBM Surges, in ‘Bucking Bronco’ High Performance Server Market

September 27, 2018

Riding healthy U.S. and global economies, strong demand for AI-capable hardware and other tailwind trends, the high performance computing server market jumped 28 percent in the second quarter 2018 to $3.7 billion, up from $2.9 billion for the same period last year, according to industry analyst firm Hyperion Research. Read more…

By Doug Black

Intel Announces Cooper Lake, Advances AI Strategy

August 9, 2018

Intel's chief datacenter exec Navin Shenoy kicked off the company's Data-Centric Innovation Summit Wednesday, the day-long program devoted to Intel's datacenter Read more…

By Tiffany Trader

GPUs Power Five of World’s Top Seven Supercomputers

June 25, 2018

The top 10 echelon of the newly minted Top500 list boasts three powerful new systems with one common engine: the Nvidia Volta V100 general-purpose graphics proc Read more…

By Tiffany Trader

Germany Celebrates Launch of Two Fastest Supercomputers

September 26, 2018

The new high-performance computer SuperMUC-NG at the Leibniz Supercomputing Center (LRZ) in Garching is the fastest computer in Germany and one of the fastest i Read more…

By Tiffany Trader

MLPerf – Will New Machine Learning Benchmark Help Propel AI Forward?

May 2, 2018

Let the AI benchmarking wars begin. Today, a diverse group from academia and industry – Google, Baidu, Intel, AMD, Harvard, and Stanford among them – releas Read more…

By John Russell

Aerodynamic Simulation Reveals Best Position in a Peloton of Cyclists

July 5, 2018

Eindhoven University of Technology (TU/e) and KU Leuven research group conducts the largest numerical simulation ever done in the sport industry and cycling discipline. The goal was to understand the aerodynamic interactions in the peloton, i.e., the main pack of cyclists in a race. Read more…

Houston to Field Massive, ‘Geophysically Configured’ Cloud Supercomputer

October 11, 2018

Based on some news stories out today, one might get the impression that the next system to crack number one on the Top500 would be an industrial oil and gas mon 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