Remote Direct Memory Access (RDMA) over IP Problem Statement

Status of This Memo

This memo provides information for the Internet community. It does not specify an Internet standard of any kind. Distribution of this memo is unlimited.

Copyright Notice

Copyright © The Internet Society (2005).

Abstract

Overhead due to the movement of user data in the end-system network I/O processing path at high speeds is significant, and has limited the use of Internet protocols in interconnection networks, and the Internet itself -- especially where high bandwidth, low latency, and/or low overhead are required by the hosted application.

This document examines this overhead, and addresses an architectural, IP-based "copy avoidance" solution for its elimination, by enabling Remote Direct Memory Access (RDMA).

Table of Contents

   1. Introduction ....................................................2
   2. The High Cost of Data Movement Operations in Network I/O ........4
      2.1. Copy avoidance improves processing overhead. ...............5
   3. Memory bandwidth is the root cause of the problem. ..............6
   4. High copy overhead is problematic for many key Internet
      applications. ...................................................8
   5. Copy Avoidance Techniques ......................................10
      5.1. A Conceptual Framework: DDP and RDMA ......................11
   6. Conclusions ....................................................12
   7. Security Considerations ........................................12
   8. Terminology ....................................................14
   9. Acknowledgements ...............................................14
   10. Informative References ........................................15

1. Introduction

This document considers the problem of high host processing overhead associated with the movement of user data to and from the network interface under high speed conditions. This problem is often referred to as the "I/O bottleneck" [CT90]. More specifically, the source of high overhead that is of interest here is data movement operations, i.e., copying. The throughput of a system may therefore be limited by the overhead of this copying. This issue is not to be confused with TCP offload, which is not addressed here. High speed refers to conditions where the network link speed is high, relative to the bandwidths of the host CPU and memory. With today's computer systems, one Gigabit per second (Gbits/s) and over is considered high speed.

High costs associated with copying are an issue primarily for large scale systems. Although smaller systems such as rack-mounted PCs and small workstations would benefit from a reduction in copying overhead, the benefit to smaller machines will be primarily in the next few years as they scale the amount of bandwidth they handle. Today, it is large system machines with high bandwidth feeds, usually multiprocessors and clusters, that are adversely affected by copying overhead. Examples of such machines include all varieties of servers: database servers, storage servers, application servers for transaction processing, for e-commerce, and web serving, content distribution, video distribution, backups, data mining and decision support, and scientific computing.

Note that such servers almost exclusively service many concurrent sessions (transport connections), which, in aggregate, are responsible for > 1 Gbits/s of communication. Nonetheless, the cost of copying overhead for a particular load is the same whether from few or many sessions.

The I/O bottleneck, and the role of data movement operations, have been widely studied in research and industry over the last approximately 14 years, and we draw freely on these results. Historically, the I/O bottleneck has received attention whenever new networking technology has substantially increased line rates: 100 Megabit per second (Mbits/s) Fast Ethernet and Fibre Distributed Data Interface [FDDI], 155 Mbits/s Asynchronous Transfer Mode [ATM], 1 Gbits/s Ethernet. In earlier speed transitions, the availability of memory bandwidth allowed the I/O bottleneck issue to be deferred. Now however, this is no longer the case. While the I/O problem is significant at 1 Gbits/s, it is the introduction of 10 Gbits/s Ethernet which is motivating an upsurge of activity in industry and research [IB, VI, CGY01, Ma02, MAF+02].

Because of high overhead of end-host processing in current implementations, the TCP/IP protocol stack is not used for high speed transfer. Instead, special purpose network fabrics, using a technology generally known as Remote Direct Memory Access (RDMA), have been developed and are widely used. RDMA is a set of mechanisms that allow the network adapter, under control of the application, to steer data directly into and out of application buffers. Examples of such interconnection fabrics include Fibre Channel [FIBRE] for block storage transfer, Virtual Interface Architecture [VI] for database clusters, and Infiniband [IB], Compaq Servernet [SRVNET], and Quadrics [QUAD] for System Area Networks. These link level technologies limit application scaling in both distance and size, meaning that the number of nodes cannot be arbitrarily large.

This problem statement substantiates the claim that in network I/O processing, high overhead results from data movement operations, specifically copying; and that copy avoidance significantly decreases this processing overhead. It describes when and why the high processing overheads occur, explains why the overhead is problematic, and points out which applications are most affected.

The document goes on to discuss why the problem is relevant to the Internet and to Internet-based applications. Applications that store, manage, and distribute the information of the Internet are well suited to applying the copy avoidance solution. They will benefit by avoiding high processing overheads, which removes limits to the available scaling of tiered end-systems. Copy avoidance also eliminates latency for these systems, which can further benefit effective distributed processing.

In addition, this document introduces an architectural approach to solving the problem, which is developed in detail in [BT05]. It also discusses how the proposed technology may introduce security concerns and how they should be addressed.

Finally, this document includes a Terminology section to aid as a reference for several new terms introduced by RDMA.

2. The High Cost of Data Movement Operations in Network I/O

A wealth of data from research and industry shows that copying is responsible for substantial amounts of processing overhead. It further shows that even in carefully implemented systems, eliminating copies significantly reduces the overhead, as referenced below.

Clark et al. [CJRS89] in 1989 shows that TCP [Po81] overhead processing is attributable to both operating system costs (such as interrupts, context switches, process management, buffer management, timer management) and the costs associated with processing individual bytes (specifically, computing the checksum and moving data in memory). They found that moving data in memory is the more important of the costs, and their experiments show that memory bandwidth is the greatest source of limitation. In the data presented [CJRS89], 64% of the measured microsecond overhead was attributable to data touching operations, and 48% was accounted for by copying. The system measured Berkeley TCP on a Sun-3/60 using 1460 Byte Ethernet packets.

In a well-implemented system, copying can occur between the network interface and the kernel, and between the kernel and application buffers; there are two copies, each of which are two memory bus crossings, for read and write. Although in certain circumstances it is possible to do better, usually two copies are required on receive.

Subsequent work has consistently shown the same phenomenon as the earlier Clark study. A number of studies report results that data- touching operations, checksumming and data movement, dominate the processing costs for messages longer than 128 Bytes [BS96, CGY01, Ch96, CJRS89, DAPP93, KP96]. For smaller sized messages, per-packet overheads dominate [KP96, CGY01].

The percentage of overhead due to data-touching operations increases with packet size, since time spent on per-byte operations scales linearly with message size [KP96]. For example, Chu [Ch96] reported substantial per-byte latency costs as a percentage of total networking software costs for an MTU size packet on a SPARCstation/20 running memory-to-memory TCP tests over networks with 3 different MTU sizes. The percentage of total software costs attributable to per-byte operations were:

      1500 Byte Ethernet 18-25%
      4352 Byte FDDI     35-50%
      9180 Byte ATM      55-65%

Although many studies report results for data-touching operations, including checksumming and data movement together, much work has focused just on copying [BS96, Br99, Ch96, TK95]. For example, [KP96] reports results that separate processing times for checksum from data movement operations. For the 1500 Byte Ethernet size, 20% of total processing overhead time is attributable to copying. The study used 2 DECstations 5000/200 connected by an FDDI network. (In this study, checksum accounts for 30% of the processing time.)

2.1. Copy avoidance improves processing overhead.

A number of studies show that eliminating copies substantially reduces overhead. For example, results from copy-avoidance in the IO-Lite system [PDZ99], which aimed at improving web server performance, show a throughput increase of 43% over an optimized web server, and 137% improvement over an Apache server. The system was implemented in a 4.4BSD-derived UNIX kernel, and the experiments used a server system based on a 333MHz Pentium II PC connected to a switched 100 Mbits/s Fast Ethernet.

There are many other examples where elimination of copying using a variety of different approaches showed significant improvement in system performance [CFF+94, DP93, EBBV95, KSZ95, TK95, Wa97]. We will discuss the results of one of these studies in detail in order to clarify the significant degree of improvement produced by copy avoidance [Ch02].

Recent work by Chase et al. [CGY01], measuring CPU utilization, shows that avoiding copies reduces CPU time spent on data access from 24% to 15% at 370 Mbits/s for a 32 KBytes MTU using an AlphaStation XP1000 and a Myrinet adapter [BCF+95]. This is an absolute improvement of 9% due to copy avoidance.

The total CPU utilization was 35%, with data access accounting for 24%. Thus, the relative importance of reducing copies is 26%. At 370 Mbits/s, the system is not very heavily loaded. The relative improvement in achievable bandwidth is 34%. This is the improvement we would see if copy avoidance were added when the machine was saturated by network I/O.

Note that improvement from the optimization becomes more important if the overhead it targets is a larger share of the total cost. This is what happens if other sources of overhead, such as checksumming, are eliminated. In [CGY01], after removing checksum overhead, copy avoidance reduces CPU utilization from 26% to 10%. This is a 16% absolute reduction, a 61% relative reduction, and a 160% relative improvement in achievable bandwidth.

In fact, today's network interface hardware commonly offloads the checksum, which removes the other source of per-byte overhead. They also coalesce interrupts to reduce per-packet costs. Thus, today copying costs account for a relatively larger part of CPU utilization than previously, and therefore relatively more benefit is to be gained in reducing them. (Of course this argument would be specious if the amount of overhead were insignificant, but it has been shown to be substantial. [BS96, Br99, Ch96, KP96, TK95])

3. Memory bandwidth is the root cause of the problem.

Data movement operations are expensive because memory bandwidth is scarce relative to network bandwidth and CPU bandwidth [PAC+97]. This trend existed in the past and is expected to continue into the future [HP97, STREAM], especially in large multiprocessor systems.

With copies crossing the bus twice per copy, network processing overhead is high whenever network bandwidth is large in comparison to CPU and memory bandwidths. Generally, with today's end-systems, the effects are observable at network speeds over 1 Gbits/s. In fact, with multiple bus crossings it is possible to see the bus bandwidth being the limiting factor for throughput. This prevents such an end-system from simultaneously achieving full network bandwidth and full application performance.

A common question is whether an increase in CPU processing power alleviates the problem of high processing costs of network I/O. The answer is no, it is the memory bandwidth that is the issue. Faster CPUs do not help if the CPU spends most of its time waiting for memory [CGY01].

The widening gap between microprocessor performance and memory performance has long been a widely recognized and well-understood problem [PAC+97]. Hennessy [HP97] shows microprocessor performance grew from 1980-1998 at 60% per year, while the access time to DRAM improved at 10% per year, giving rise to an increasing "processor- memory performance gap".

Another source of relevant data is the STREAM Benchmark Reference Information website, which provides information on the STREAM benchmark [STREAM]. The benchmark is a simple synthetic benchmark program that measures sustainable memory bandwidth (in MBytes/s) and the corresponding computation rate for simple vector kernels measured in MFLOPS. The website tracks information on sustainable memory bandwidth for hundreds of machines and all major vendors.

Results show measured system performance statistics. Processing performance from 1985-2001 increased at 50% per year on average, and sustainable memory bandwidth from 1975 to 2001 increased at 35% per year, on average, over all the systems measured. A similar 15% per year lead of processing bandwidth over memory bandwidth shows up in another statistic, machine balance [Mc95], a measure of the relative rate of CPU to memory bandwidth (FLOPS/cycle) / (sustained memory ops/cycle) [STREAM].

Network bandwidth has been increasing about 10-fold roughly every 8 years, which is a 40% per year growth rate.

   A typical example illustrates that the memory bandwidth compares
   unfavorably with link speed.  The STREAM benchmark shows that a
   modern uniprocessor PC, for example the 1.2 GHz Athlon in 2001, will
   move the data 3 times in doing a receive operation: once for the
   network interface to deposit the data in memory, and twice for the
   CPU to copy the data.  With 1 GBytes/s of memory bandwidth, meaning
   one read or one write, the machine could handle approximately 2.67
   Gbits/s of network bandwidth, one third the copy bandwidth.  But this
   assumes 100% utilization, which is not possible, and more importantly
   the machine would be totally consumed!  (A rule of thumb for
   databases is that 20% of the machine should be required to service
   I/O, leaving 80% for the database application.  And, the less, the
   better.)

In 2001, 1 Gbits/s links were common. An application server may typically have two 1 Gbits/s connections: one connection backend to a storage server and one front-end, say for serving HTTP [FGM+99]. Thus, the communications could use 2 Gbits/s. In our typical example, the machine could handle 2.7 Gbits/s at its theoretical maximum while doing nothing else. This means that the machine basically could not keep up with the communication demands in 2001; with the relative growth trends, the situation only gets worse.

4. High copy overhead is problematic for many key Internet

applications.

If a significant portion of resources on an application machine is consumed in network I/O rather than in application processing, it makes it difficult for the application to scale, i.e., to handle more clients, to offer more services.

Several years ago the most affected applications were streaming multimedia, parallel file systems, and supercomputing on clusters [BS96]. In addition, today the applications that suffer from copying overhead are more central in Internet computing -- they store, manage, and distribute the information of the Internet and the enterprise. They include database applications doing transaction processing, e-commerce, web serving, decision support, content distribution, video distribution, and backups. Clusters are typically used for this category of application, since they have advantages of availability and scalability.

Today these applications, which provide and manage Internet and corporate information, are typically run in data centers that are organized into three logical tiers. One tier is typically a set of web servers connecting to the WAN. The second tier is a set of application servers that run the specific applications usually on more powerful machines, and the third tier is backend databases. Physically, the first two tiers -- web server and application server -- are usually combined [Pi01]. For example, an e-commerce server communicates with a database server and with a customer site, or a content distribution server connects to a server farm, or an OLTP server connects to a database and a customer site.

When network I/O uses too much memory bandwidth, performance on network paths between tiers can suffer. (There might also be performance issues on Storage Area Network paths used either by the database tier or the application tier.) The high overhead from network-related memory copies diverts system resources from other application processing. It also can create bottlenecks that limit total system performance.

There is high motivation to maximize the processing capacity of each CPU because scaling by adding CPUs, one way or another, has drawbacks. For example, adding CPUs to a multiprocessor will not necessarily help because a multiprocessor improves performance only when the memory bus has additional bandwidth to spare. Clustering can add additional complexity to handling the applications.

In order to scale a cluster or multiprocessor system, one must proportionately scale the interconnect bandwidth. Interconnect bandwidth governs the performance of communication-intensive parallel applications; if this (often expressed in terms of "bisection bandwidth") is too low, adding additional processors cannot improve system throughput. Interconnect latency can also limit the performance of applications that frequently share data between processors.

So, excessive overheads on network paths in a "scalable" system both can require the use of more processors than optimal, and can reduce the marginal utility of those additional processors.

Copy avoidance scales a machine upwards by removing at least two- thirds of the bus bandwidth load from the "very best" 1-copy (on receive) implementations, and removes at least 80% of the bandwidth overhead from the 2-copy implementations.

The removal of bus bandwidth requirements, in turn, removes bottlenecks from the network processing path and increases the throughput of the machine. On a machine with limited bus bandwidth, the advantages of removing this load is immediately evident, as the host can attain full network bandwidth. Even on a machine with bus bandwidth adequate to sustain full network bandwidth, removal of bus bandwidth load serves to increase the availability of the machine for the processing of user applications, in some cases dramatically.

An example showing poor performance with copies and improved scaling with copy avoidance is illustrative. The IO-Lite work [PDZ99] shows higher server throughput servicing more clients using a zero-copy system. In an experiment designed to mimic real world web conditions by simulating the effect of TCP WAN connections on the server, the performance of 3 servers was compared. One server was Apache, another was an optimized server called Flash, and the third was the Flash server running IO-Lite, called Flash-Lite with zero copy. The measurement was of throughput in requests/second as a function of the number of slow background clients that could be served. As the table shows, Flash-Lite has better throughput, especially as the number of clients increases.

              Apache              Flash         Flash-Lite
              ------              -----         ----------
   #Clients   Throughput reqs/s   Throughput    Throughput
   
   0          520                 610           890
   16         390                 490           890
   32         360                 490           850
   64         360                 490           890
   128        310                 450           880
   256        310                 440           820

Traditional Web servers (which mostly send data and can keep most of their content in the file cache) are not the worst case for copy overhead. Web proxies (which often receive as much data as they send) and complex Web servers based on System Area Networks or multi-tier systems will suffer more from copy overheads than in the example above.

5. Copy Avoidance Techniques

There have been extensive research investigation and industry experience with two main alternative approaches to eliminating data movement overhead, often along with improving other Operating System processing costs. In one approach, hardware and/or software changes within a single host reduce processing costs. In another approach, memory-to-memory networking [MAF+02], the exchange of explicit data placement information between hosts allows them to reduce processing costs.

The single host approaches range from new hardware and software architectures [KSZ95, Wa97, DWB+93] to new or modified software systems [BS96, Ch96, TK95, DP93, PDZ99]. In the approach based on using a networking protocol to exchange information, the network adapter, under control of the application, places data directly into and out of application buffers, reducing the need for data movement. Commonly this approach is called RDMA, Remote Direct Memory Access.

As discussed below, research and industry experience has shown that copy avoidance techniques within the receiver processing path alone have proven to be problematic. The research special purpose host adapter systems had good performance and can be seen as precursors for the commercial RDMA-based adapters [KSZ95, DWB+93]. In software, many implementations have successfully achieved zero-copy transmit, but few have accomplished zero-copy receive. And those that have done so make strict alignment and no-touch requirements on the application, greatly reducing the portability and usefulness of the implementation.

In contrast, experience has proven satisfactory with memory-to-memory systems that permit RDMA; performance has been good and there have not been system or networking difficulties. RDMA is a single solution. Once implemented, it can be used with any OS and machine architecture, and it does not need to be revised when either of these are changed.

In early work, one goal of the software approaches was to show that TCP could go faster with appropriate OS support [CJRS89, CFF+94]. While this goal was achieved, further investigation and experience showed that, though possible to craft software solutions, specific system optimizations have been complex, fragile, extremely interdependent with other system parameters in complex ways, and often of only marginal improvement [CFF+94, CGY01, Ch96, DAPP93, KSZ95, PDZ99]. The network I/O system interacts with other aspects of the Operating System such as machine architecture and file I/O, and disk I/O [Br99, Ch96, DP93].

For example, the Solaris Zero-Copy TCP work [Ch96], which relies on page remapping, shows that the results are highly interdependent with other systems, such as the file system, and that the particular optimizations are specific for particular architectures, meaning that for each variation in architecture, optimizations must be re-crafted [Ch96].

With RDMA, application I/O buffers are mapped directly, and the authorized peer may access it without incurring additional processing overhead. When RDMA is implemented in hardware, arbitrary data movement can be performed without involving the host CPU at all.

A number of research projects and industry products have been based on the memory-to-memory approach to copy avoidance. These include U-Net [EBBV95], SHRIMP [BLA+94], Hamlyn [BJM+96], Infiniband [IB], Winsock Direct [Pi01]. Several memory-to-memory systems have been widely used and have generally been found to be robust, to have good performance, and to be relatively simple to implement. These include VI [VI], Myrinet [BCF+95], Quadrics [QUAD], Compaq/Tandem Servernet [SRVNET]. Networks based on these memory-to-memory architectures have been used widely in scientific applications and in data centers for block storage, file system access, and transaction processing.

By exporting direct memory access "across the wire", applications may direct the network stack to manage all data directly from application buffers. A large and growing class that takes advantage of such capabilities of applications has already emerged. It includes all the major databases, as well as network protocols such as Sockets Direct [SDP].

5.1. A Conceptual Framework: DDP and RDMA

An RDMA solution can be usefully viewed as being comprised of two distinct components: "direct data placement (DDP)" and "remote direct memory access (RDMA) semantics". They are distinct in purpose and also in practice -- they may be implemented as separate protocols.

The more fundamental of the two is the direct data placement facility. This is the means by which memory is exposed to the remote peer in an appropriate fashion, and the means by which the peer may access it, for instance, reading and writing.

The RDMA control functions are semantically layered atop direct data placement. Included are operations that provide "control" features, such as connection and termination, and the ordering of operations and signaling their completions. A "send" facility is provided.

While the functions (and potentially protocols) are distinct, historically both aspects taken together have been referred to as "RDMA". The facilities of direct data placement are useful in and of themselves, and may be employed by other upper layer protocols to facilitate data transfer. Therefore, it is often useful to refer to DDP as the data placement functionality and RDMA as the control aspect.

[BT05] develops an architecture for DDP and RDMA atop the Internet Protocol Suite, and is a companion document to this problem statement.

6. Conclusions

This Problem Statement concludes that an IP-based, general solution for reducing processing overhead in end-hosts is desirable.

It has shown that high overhead of the processing of network data leads to end-host bottlenecks. These bottlenecks are in large part attributable to the copying of data. The bus bandwidth of machines has historically been limited, and the bandwidth of high-speed interconnects taxes it heavily.

An architectural solution to alleviate these bottlenecks best satisfies the issue. Further, the high speed of today's interconnects and the deployment of these hosts on Internet Protocol-based networks leads to the desirability of layering such a solution on the Internet Protocol Suite. The architecture described in [BT05] is such a proposal.

7. Security Considerations

Solutions to the problem of reducing copying overhead in high bandwidth transfers may introduce new security concerns. Any proposed solution must be analyzed for security vulnerabilities and any such vulnerabilities addressed. Potential security weaknesses -- due to resource issues that might lead to denial-of-service attacks, overwrites and other concurrent operations, the ordering of completions as required by the RDMA protocol, the granularity of transfer, and any other identified vulnerabilities -- need to be examined, described, and an adequate resolution to them found.

Layered atop Internet transport protocols, the RDMA protocols will gain leverage from and must permit integration with Internet security standards, such as IPsec and TLS [IPSEC, TLS]. However, there may be implementation ramifications for certain security approaches with respect to RDMA, due to its copy avoidance.

IPsec, operating to secure the connection on a packet-by-packet basis, seems to be a natural fit to securing RDMA placement, which operates in conjunction with transport. Because RDMA enables an implementation to avoid buffering, it is preferable to perform all applicable security protection prior to processing of each segment by the transport and RDMA layers. Such a layering enables the most efficient secure RDMA implementation.

The TLS record protocol, on the other hand, is layered on top of reliable transports and cannot provide such security assurance until an entire record is available, which may require the buffering and/or assembly of several distinct messages prior to TLS processing. This defers RDMA processing and introduces overheads that RDMA is designed to avoid. Therefore, TLS is viewed as potentially a less natural fit for protecting the RDMA protocols.

It is necessary to guarantee properties such as confidentiality, integrity, and authentication on an RDMA communications channel. However, these properties cannot defend against all attacks from properly authenticated peers, which might be malicious, compromised, or buggy. Therefore, the RDMA design must address protection against such attacks. For example, an RDMA peer should not be able to read or write memory regions without prior consent.

Further, it must not be possible to evade memory consistency checks at the recipient. The RDMA design must allow the recipient to rely on its consistent memory contents by explicitly controlling peer access to memory regions at appropriate times.

Peer connections that do not pass authentication and authorization checks by upper layers must not be permitted to begin processing in RDMA mode with an inappropriate endpoint. Once associated, peer accesses to memory regions must be authenticated and made subject to authorization checks in the context of the association and connection on which they are to be performed, prior to any transfer operation or data being accessed.

The RDMA protocols must ensure that these region protections be under strict application control. Remote access to local memory by a network peer is particularly important in the Internet context, where such access can be exported globally.

8. Terminology

This section contains general terminology definitions for this document and for Remote Direct Memory Access in general.

Remote Direct Memory Access (RDMA)

A method of accessing memory on a remote system in which the

local system specifies the location of the data to be

transferred.

RDMA Protocol

A protocol that supports RDMA Operations to transfer data between systems.

Fabric

The collection of links, switches, and routers that connect a set of systems.

Storage Area Network (SAN)

A network where disks, tapes, and other storage devices are made available to one or more end-systems via a fabric.

System Area Network

A network where clustered systems share services, such as

storage and interprocess communication, via a fabric.

Fibre Channel (FC)

An ANSI standard link layer with associated protocols, typically used to implement Storage Area Networks. [FIBRE]

   Virtual Interface Architecture (VI, VIA)
        An RDMA interface definition developed by an industry group and
        implemented with a variety of differing wire protocols. [VI]

Infiniband (IB)

An RDMA interface, protocol suite and link layer specification defined by an industry trade association. [IB]

9. Acknowledgements

Jeff Chase generously provided many useful insights and information. Thanks to Jim Pinkerton for many helpful discussions.

10. Informative References

   [ATM]      The ATM Forum, "Asynchronous Transfer Mode Physical Layer
              Specification" af-phy-0015.000, etc.  available from
              http://www.atmforum.com/standards/approved.html.
   
   [BCF+95]   N. J. Boden, D. Cohen, R. E. Felderman, A. E. Kulawik, C.
              L. Seitz, J. N. Seizovic, and W. Su. "Myrinet - A
              gigabit-per-second local-area network", IEEE Micro,
              February 1995.
   
   [BJM+96]   G. Buzzard, D. Jacobson, M. Mackey, S. Marovich, J.
              Wilkes, "An implementation of the Hamlyn send-managed
              interface architecture", in Proceedings of the Second
              Symposium on Operating Systems Design and Implementation,
              USENIX Assoc., October 1996.
   
   [BLA+94]   M. A. Blumrich, K. Li, R. Alpert, C. Dubnicki, E. W.
              Felten, "A virtual memory mapped network interface for the
              SHRIMP multicomputer", in Proceedings of the 21st Annual
              Symposium on Computer Architecture, April 1994, pp. 142-
              153.
   
   [Br99]     J. C. Brustoloni, "Interoperation of copy avoidance in
              network and file I/O", Proceedings of IEEE Infocom, 1999,
              pp. 534-542.
   
   [BS96]     J. C. Brustoloni, P. Steenkiste, "Effects of buffering
              semantics on I/O performance", Proceedings OSDI'96,
              USENIX, Seattle, WA October 1996, pp. 277-291.
   
   [BT05]     Bailey, S. and T. Talpey, "The Architecture of Direct Data
              Placement (DDP) And Remote Direct Memory Access (RDMA) On
              Internet Protocols", RFC 4296, December 2005.
   
   [CFF+94]   C-H Chang, D. Flower, J. Forecast, H. Gray, B. Hawe, A.
              Nadkarni, K. K. Ramakrishnan, U. Shikarpur, K. Wilde,
              "High-performance TCP/IP and UDP/IP networking in DEC
              OSF/1 for Alpha AXP",  Proceedings of the 3rd IEEE
              Symposium on High Performance Distributed Computing,
              August 1994, pp. 36-42.
   
   [CGY01]    J. S. Chase, A. J. Gallatin, and K. G. Yocum, "End system
              optimizations for high-speed TCP", IEEE Communications
              Magazine, Volume: 39, Issue: 4 , April 2001, pp 68-74.
              http://www.cs.duke.edu/ari/publications/end-
              system.{ps,pdf}.
   
   [Ch96]     H.K. Chu, "Zero-copy TCP in Solaris", Proc. of the USENIX
              1996 Annual Technical Conference, San Diego, CA, January
              1996.
   
   [Ch02]     Jeffrey Chase, Personal communication.
   
   [CJRS89]   D. D. Clark, V. Jacobson, J. Romkey, H. Salwen, "An
              analysis of TCP processing overhead", IEEE Communications
              Magazine, volume:  27, Issue: 6, June 1989, pp 23-29.
   
   [CT90]     D. D. Clark, D. Tennenhouse, "Architectural considerations
              for a new generation of protocols", Proceedings of the ACM
              SIGCOMM Conference, 1990.
   
   [DAPP93]   P. Druschel, M. B. Abbott, M. A. Pagels, L. L. Peterson,
              "Network subsystem design", IEEE Network, July 1993, pp.
              8-17.
   
   [DP93]     P. Druschel, L. L. Peterson, "Fbufs: a high-bandwidth
              cross-domain transfer facility", Proceedings of the 14th
              ACM Symposium of Operating Systems Principles, December
              1993.
   
   [DWB+93]   C. Dalton, G. Watson, D. Banks, C. Calamvokis, A. Edwards,
              J. Lumley, "Afterburner: architectural support for high-
              performance protocols", Technical Report, HP Laboratories
              Bristol, HPL-93-46, July 1993.
   
   [EBBV95]   T. von Eicken, A. Basu, V. Buch, and W. Vogels, "U-Net: A
              user-level network interface for parallel and distributed
              computing", Proc. of the 15th ACM Symposium on Operating
              Systems Principles, Copper Mountain, Colorado, December
              3-6, 1995.
   
   [FDDI]     International Standards Organization, "Fibre Distributed
              Data Interface", ISO/IEC 9314, committee drafts available
              from http://www.iso.org.
   
   [FGM+99]   Fielding,  R., Gettys, J., Mogul, J., Frystyk, H.,
              Masinter, L., Leach, P., and T. Berners-Lee, "Hypertext
              Transfer Protocol -- HTTP/1.1", RFC 2616, June 1999.
   
   [FIBRE]    ANSI Technical Committee T10, "Fibre Channel Protocol
              (FCP)" (and as revised and updated), ANSI X3.269:1996
              [R2001], committee draft available from
              http://www.t10.org/drafts.htm#FibreChannel
   
   [HP97]     J. L. Hennessy, D. A. Patterson, Computer Organization and
              Design, 2nd Edition, San Francisco: Morgan Kaufmann
              Publishers, 1997.
   
   [IB]       InfiniBand Trade Association, "InfiniBand Architecture
              Specification, Volumes 1 and 2", Release 1.1, November
              2002, available from http://www.infinibandta.org/specs.
   
   [IPSEC]    Kent, S. and R. Atkinson, "Security Architecture for the
              Internet Protocol", RFC 2401, November 1998.
   
   [KP96]     J. Kay, J. Pasquale, "Profiling and reducing processing
              overheads in TCP/IP", IEEE/ACM Transactions on Networking,
              Vol 4, No. 6, pp.817-828, December 1996.
   
   [KSZ95]    K. Kleinpaste, P. Steenkiste, B. Zill, "Software support
              for outboard buffering and checksumming", SIGCOMM'95.
   
   [Ma02]     K. Magoutis, "Design and Implementation of a Direct Access
              File System (DAFS) Kernel Server for FreeBSD", in
              Proceedings of USENIX BSDCon 2002 Conference, San
              Francisco, CA, February 11-14, 2002.
   
   [MAF+02]   K. Magoutis, S. Addetia, A. Fedorova, M.  I. Seltzer, J.
              S. Chase, D. Gallatin, R. Kisley, R. Wickremesinghe, E.
              Gabber, "Structure and Performance of the Direct Access
              File System (DAFS)", in Proceedings of the 2002 USENIX
              Annual Technical Conference, Monterey, CA, June 9-14,
              2002.
   
   [Mc95]     J. D. McCalpin, "A Survey of memory bandwidth and machine
              balance in current high performance computers", IEEE TCCA
              Newsletter, December 1995.
   
   [PAC+97]   D. Patterson, T. Anderson, N. Cardwell, R. Fromm, K.
              Keeton, C. Kozyrakis, R. Thomas, K. Yelick , "A case for
              intelligient RAM: IRAM", IEEE Micro, April 1997.
   
   [PDZ99]    V. S. Pai, P. Druschel, W. Zwaenepoel, "IO-Lite: a unified
              I/O buffering and caching system", Proc. of the 3rd
              Symposium on Operating Systems Design and Implementation,
              New Orleans, LA, February 1999.
   
   [Pi01]     J. Pinkerton, "Winsock Direct: The Value of System Area
              Networks", May 2001, available from
              http://www.microsoft.com/windows2000/techinfo/
              howitworks/communications/winsock.asp.
   
   [Po81]     Postel, J., "Transmission Control Protocol", STD 7, RFC
              793, September 1981.
   
   [QUAD]     Quadrics Ltd., Quadrics QSNet product information,
              available from
              http://www.quadrics.com/website/pages/02qsn.html.
   
   [SDP]      InfiniBand Trade Association, "Sockets Direct Protocol
              v1.0", Annex A of InfiniBand Architecture Specification
              Volume 1, Release 1.1, November 2002, available from
              http://www.infinibandta.org/specs.
   
   [SRVNET]   R. Horst, "TNet: A reliable system area network", IEEE
              Micro, pp. 37-45, February 1995.
   
   [STREAM]   J. D. McAlpin, The STREAM Benchmark Reference Information,
              http://www.cs.virginia.edu/stream/.
   
   [TK95]     M. N. Thadani, Y. A. Khalidi, "An efficient zero-copy I/O
              framework for UNIX", Technical Report, SMLI TR-95-39, May
              1995.
   
   [TLS]      Dierks, T. and C. Allen, "The TLS Protocol Version 1.0",
              RFC 2246, January 1999.
   
   [VI]       D. Cameron and G. Regnier, "The Virtual Interface
              Architecture", ISBN 0971288704, Intel Press, April 2002,
              more info at http://www.intel.com/intelpress/via/.
   
   [Wa97]     J. R. Walsh, "DART: Fast application-level networking via
              data-copy avoidance", IEEE Network, July/August 1997, pp.
              28-38.

Authors' Addresses

   Stephen Bailey
   Sandburst Corporation
   600 Federal Street
   Andover, MA  01810 USA
   
   Phone: +1 978 689 1614
   EMail: steph@sandburst.com
   
   Jeffrey C. Mogul
   HP Labs
   Hewlett-Packard Company
   1501 Page Mill Road, MS 1117
   Palo Alto, CA  94304 USA
   
   Phone: +1 650 857 2206 (EMail preferred)
   EMail: JeffMogul@acm.org
   
   Allyn Romanow
   Cisco Systems, Inc.
   170 W. Tasman Drive
   San Jose, CA  95134 USA
   
   Phone: +1 408 525 8836
   EMail: allyn@cisco.com
   
   Tom Talpey
   Network Appliance
   1601 Trapelo Road
   Waltham, MA  02451 USA
   
   Phone: +1 781 768 5329
   EMail: thomas.talpey@netapp.com

Full Copyright Statement

Copyright © The Internet Society (2005).

This document is subject to the rights, licenses and restrictions contained in BCP 78, and except as set forth therein, the authors retain all their rights.

This document and the information contained herein are provided on an "AS IS" basis and THE CONTRIBUTOR, THE ORGANIZATION HE/SHE REPRESENTS OR IS SPONSORED BY (IF ANY), THE INTERNET SOCIETY AND THE INTERNET ENGINEERING TASK FORCE DISCLAIM ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.

Intellectual Property

The IETF takes no position regarding the validity or scope of any Intellectual Property Rights or other rights that might be claimed to pertain to the implementation or use of the technology described in this document or the extent to which any license under such rights might or might not be available; nor does it represent that it has made any independent effort to identify any such rights. Information on the procedures with respect to rights in RFC documents can be found in BCP 78 and BCP 79.

Copies of IPR disclosures made to the IETF Secretariat and any assurances of licenses to be made available, or the result of an attempt made to obtain a general license or permission for the use of such proprietary rights by implementers or users of this specification can be obtained from the IETF on-line IPR repository at http://www.ietf.org/ipr.

The IETF invites any interested party to bring to its attention any copyrights, patents or patent applications, or other proprietary rights that may cover technology that may be required to implement this standard. Please address the information to the IETF at ietf- ipr@ietf.org.

Acknowledgement

Funding for the RFC Editor function is currently provided by the Internet Society.