Chris Dalton, Greg Watson, David Banks, Costas Calamvokis, Aled Edwards, and John Lumley. Afterburner. IEEE Network, 7(4):36--43, July 1993.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....but what performance bene t, if any, do they provide Does transmitfile( or send file( show any improvement over the already available mmap( and writev( system calls per byte optimizations. It is well known that data touching operations, such as copying and checksumming, are expensive [5, 11, 13, 23]. BSD derived Unix operating systems [24] use di erent bu ering mechanisms in the le system and the networking code, forcing data to be copied when it is moved from one The authors are with the IBM T.J. Watson Research Center. subsystem to another. How well can we approximate a zero copy ....

Chris Dalton, Greg Watson, David Banks, Costas Clamvokis, Aled Edwards, and John Lumley. Afterburner. IEEE Network, 11(2):36-43, July 1993.

....reconstruction within the selfsecuring NI software. Although this work is redundant with respect to the host s network stack, it allows self securing NIs to be deployed with no client software modification. For NIs that offload higher level protocols (e.g. IP security or TCP) from the host [11, 13], redundant work becomes unnecessary because the only instance of the work is already within the NI. Self securing NIs enforce policies set by the network administrator, just as a centralized firewall would. In fact, administrators will configure and manage self securing NIs over the network, ....

Chris Dalton, Greg Watson, David Banks, Costas Calamvokis, Aled Edwards, and John Lumley. Afterburner. IEEE Network, 7(4):36--43, July 1993.

.... [Gi194] Studies have shown that even today, network protocols spend a significant amount of time simply copying data [Ste94] Therefore, many designs have attempted to avoid redundant copying at the application interface [DP93, RAC96, Wi192] the OS [kJC96] and the network interface [OZH 96, DWB 93, LC95, BCM94] To push the envelope of possibilities, we should consider whether it is possible to efficiently implement messaging with no extra copying, where message data are only copied out of sender s data structures into the sender s NI and from the receiver s NI to the receiver s data ....

Chris Dalton, Greg Watson, David Banks, Costas Calamvokis, Aled Edwards, and John Lumley. Afterburner. IEEE Network, pages 36-43, July 1993. 196

....context of Real Time Mach [14] based upon design principles for CPU scheduling different from ours. There has been growing interest in user level protocol implementation in recent years [11, 19] A user level TCP implementation on top of the Jetstream high speed network interface is described in [4]. U Net [20] integrates interface firmware with host software in a design that provides user level access to a network without kernel intervention. These works target high performance on the send receive path, without paying much attention to the issues of connection management, routing ....

Chris Dalton, Greg Watson, David Banks, Costas Calamvokis, Aled Edwards, and John Lumley. Afterburner. IEEE Network, 7(4):36--43, July 1993.

....frequency gives the same result as slower memory. 3 Related Work Van Jacobson proposed the WITLESS network adapter design with a large on board buffer memory [48] The WITLESS design was used by a group at HP Labs in Bristol in a series of network adapters including Medusa [18] Afterburner [28, 52] and Jetstream [38] Two more high performance network adapters were developed at about the same time. Traw and Smith [74, 75, 76] implemented an adapter for the IBM RS 6000 and Davie [29, 30, 31] implemented one for DEC workstations. The performance of TCP implementations have been extensively ....

....[60] Several researchers have reported on work with folding together the TCP or UDP checksum with the user to kernel copy in Unix. This design is often called single copy . It was presented by Jacobson as part of the WITLESS design [48] and implemented in the software for the Afterburner [18, 28] family of network adapters. A single copy design was also used by Partridge and Pink [62] when optimizing the Berkeley UDP implementation in SunOS. Implementing protocols in the application address space is not without problems. Maeda and Bershad [54] Thekkath et al. 73] Edwards and Muir ....

[Article contains additional citation context not shown here]

Chris Dalton, Greg Watson, David Banks, Costas Calamvokis, Aled Edwards, and John Lumley. Afterburner. IEEE Network, 7(4):36--43, July 1993.

.... avoidance Several projects have used page remapping and smart interface buffer allocation to accelerate processor tointerface communication, including the fbufs work at the University of Arizona [Druschel93] the Medusa FDDI interface [Lumley92, Banks93] and the follow on Afterburner project [Dalton93]. The Nectar system [Cooper90] allowed applications direct access to its communication interface memory in order to eliminate copies at the cost of all accesses being to memory in the I O space. It achieved round trip RPC latencies of 500s across a 100Mb s network. ATM network interfaces can use ....

Chris Dalton, Greg Watson, David Banks, Costas Calamvokis, Aled Edwards, and John Lumley. Afterburner. IEEE Network 7(4):36--43, July 1993.

....on the smoothness of video arrivals (Figure 18) 9 Conclusion We have proposed an architecture including several concepts designed to provide efficient support for multimedia networking. The concept of I O efficient buffers is similar to various proposals for minimizing data copies, e.g. [4], 6] and [10] Independent threads of control for network I O have been used in several systems, such as [7] We have shown how a thread of control implemented in kernel space can 1) efficiently support sends to the network by user processes, and 2) provide rate based flow control to a shared ....

Chris Dalton, Greg Watson, David Banks, Costas Calamvokis, Aled Edwards, and John Lumley. Afterburner. IEEE Network, pages 36-- 43, July 1993.

....for so called single copy protocol stack for TCP IP (i.e. only one copy operation from network interface buffer to application program buffer) a significant speedup can be achieved, especially if the packet size is large enough. They report a TCP IP throughput of up to 210 MByte s on HP 9000 730 [2]. In a summary of our tests we conclude that it is better to build large user data packets in the application layer because IP performs the fragmentation and defragmentation anyway and does so more efficiently. Especially in a heterogeneous environment the PDU size and interpacket gap for each ....

....frames to graphics adapter. Other copy operations performed by the operating system or by the X server decrease the overall system performance but are beyond our control. Other research groups are working on minimizing copy operations within the communication system [4] especially for TCP IP [2], and improving the operating system [5] to handle CM streams as efficiently as possible. 5 Summary We summarize that careful analysis of bottlenecks, provision of new algorithms such as forward error correction, and computer supported optimization of all 2 The number of SHM segments ....

Chris Dalton, Greg Watson, David Banks, Costas Calamvokis, Aled Edwards, and John Lumley. Afterburner. IEEE Network, 7(4):36--43, July 1993.

....transaction, but do they provide any performance benefit Does transmitfile( or send file( show any improvement over the already available mmap( and writev( system calls ffl per byte optimizations. It is well known that data touching operations, such as copying and checksumming, are expensive [5, 11, 13, 23]. BSD derived Unix operating systems [24] use different buffering mechanisms in the file system and the networking code, forcing data to be copied when it is moved from one subsystem to another. How well can we approximate a zero copy integrated I O architecture [31] while continuing to exploit ....

Chris Dalton, Greg Watson, David Banks, Costas Clamvokis, Aled Edwards, and John Lumley. Afterburner. IEEE Network, 11(2):36--43, July 1993.

.... of information technology [17] Even today, studies have shown that network protocols spend a significant amount of time simply copying data [49] Therefore, many designs have attempted to avoid redundant copying at the application interface [13,42,56] the OS [25] and the network interface [36,11,26,1]. To push the envelope of possibilities, we ask whether it is possible to efficiently implement messaging with no extra copying where message data are only copied out of sender s data structures into the sender s NI and from the receiver s NI to the receiver s data structures (the data should ....

Chris Dalton, Greg Watson, David Banks, Costas Calamvokis, Aled Edwards, and John Lumley. Afterburner. IEEE Network, pages 36--43, July 1993.

....were considered the key factor in performance. This led to several studies that tried to reduce the cost of software processing, e.g. 17] 28] 72] and many others. Reduced protocol costs led to research on the structure of network controllers and other host architectural effects, e.g. 5] [23], 38] 66] to name a few. The advent of a new generation of ATM networks has, in turn, resulted in many new research efforts, including ours, aimed at higher performance network communication. The work described in this thesis shares similarities with many previous studies. For example, network ....

Chris Dalton, Greg Watson, David Banks, Costas Calamvokis, Aled Edwards, and John Lumley. Afterburner. IEEE Network, 7(4):36--43, July 1993.

....packets on the smoothness of video arrivals (Figure 18) 9 Conclusion We have proposed an architecture including several concepts designed to provide efficient support for multimedia networking. The concept of I O efficient buffers is similar to various proposals for minimizing data copies, e.g. [4], 6] and [10] Independent threads of control for network I O have been used in several systems, such as [7] We have shown how a thread of control implemented in ker 0 2 4 6 8 10 12 14 0 5 10 15 20 25 30 35 Time (ms) Figure 17: Distribution of inter packet arrival times for audio with rate ....

Chris Dalton, Greg Watson, David Banks, Costas Calamvokis, Aled Edwards, and John Lumley. Afterburner. IEEE Network, pages 36--43, July 1993.

....using the CPU is often used to move data over the boundary. Copying means that the data must be transferred out of and into memory. The transfer of data in and out of memory is one of the main performance bottlenecks in the implementation of high speed communication for workstation class computers [7, 14, 9, 6, 15, 16]. The reason being that the bandwidth of the communication network is approaching that of memory and bus bandwidths in these systems. Reducing memory bandwidth utilization can substantially improve performance. As an example, Druschel and Peterson [10] measures the cost 1 to copy a memory page ....

....both in the context of operating systems [19, 13, 11] and network adapters [14, 15] During the last few years several high performance host network adapters have been designed. Some of them use DMA to transfer data to host memory [7, 18] and others utilize a large on board shared buffer memory [6, 2]. 6 Conclusions and future work We have described a communication subsystem architecture which is designed for Integrated Layer Processing and Application Layer Framing including a concrete no copy application program interface. Initial experiments show that the no copy API gives a significant ....

Chris Dalton, Greg Watson, David Banks, Costas Calamvokis, Aled Edwards, and John Lumley. Afterburner. IEEE Network, 7(4):36--43, July 1993.

....of the potential benefits of Scout paths. It is important, however, to keep in mind that the approach taken in this case study is by no means the only way paths can be exploited to improve execution speed of a system. Dynamic code generation [60] manually crafted vertically integrated code paths [56, 23], or a language based approach [14] represent a few other possibilities in this spectrum. The case study proposes and analyzes four techniques targeted at improving protocol processing. Of these techniques, the first three are path based and the last one is a compiler based technique that ....

....most can be measured. Towards this end, we collected two additional data sets. The first is a set of instruction traces that cover most of the protocol processing. The second is a set of fine grained 3 Numbers is this range have been reported in the literature for FDDI and ATM controllers [23]. 130 measurements of the execution time of the traced code. The instruction traces do not cover all of the processing since the tracing facility did not allow the tracing of interrupt handling, but other than that, the traces are complete. 4.3.3.1 Cache Statistics Using the execution traces ....

Chris Dalton, Greg Watson, David Banks, Costas Calamvokis, Aled Edwards, and John Lumley. Afterburner. IEEE Network, 7(4):35--43, July 1993.

....a timeout on message handling for protection as in FUGU. The M machine [12] receives messages with a trusted handler that has the ability to quickly forward the message body to a user thread. Memory Based Interfaces. Memory based interfaces in multicomputers [4, 7, 29, 30, 32] and workstations [9, 11, 33, 34] provide easy protection for multiprogramming if the NI also demultiplexes messages into per process buffers. Automatic hardware buffering also deals well with sinking bursts of messages and provides the lowest overhead (by avoiding the processors) when messages are not handled immediately. ....

Chris Dalton, Greg Watson, David Banks, Costas Calamvokis, Aled Edwards, and John Lumley. Afterburner. IEEE Network, pages 36-- 43, July 1993.

....control (only first packet of each picture is shown) 9 Conclusion We have proposed an architecture, and investigated various concepts and techniques, designed to support multimedia networking. The concept of I O efficient buffers is similar to various proposals for minimizing data copies, e.g. [4], 6] and [10] Independent threads of control for network I O have been used in several systems, such as [7] We have shown how a thread of control implemented in kernel space can 1) efficiently support sends to the network by user processes, and 2) provide rate based flow control to a ....

Chris Dalton, Greg Watson, David Banks, Costas Calamvokis, Aled Edwards, and John Lumley. Afterburner. IEEE Network, pages 36--43, July 1993.

No context found.

Chris Dalton, Greg Watson, David Banks, Costas Calamvokis, Aled Edwards, and John Lumley. Afterburner. IEEE Network, 7(4):36--43, July 1993.

No context found.

Chris Dalton, Greg Watson, David Banks, Costas Calamvokis, Aled Edwards, and John Lumley. Afterburner. IEEE Network, 7(4):36--43, July 1993.

No context found.

Chris Dalton, Greg Watson, David Banks, Costas Calamvokis, Aled Edwards, and John Lumley. Afterburner. IEEE Network, 7(4):36-- 43, July 1993.

No context found.

Chris Dalton, Greg Watson, David Banks, Costas Calamvokis, Aled Edwards, and John Lumley. Afterburner. IEEE Network, 7(4):35-- 43, July 1993.

No context found.

Chris Dalton, Greg Watson, David Banks, Costas Calamvokis, Aled Edwards, and John Lumley. Afterburner. IEEE Network, 7(4):36--43, July 1993.

No context found.

Chris Dalton, Greg Watson, David Banks, Costas Calamvokis, Aled Edwards, and John Lumley. Afterburner. IEEE Network, pages 36--43, July 1993.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC