Eicken, Culler, Goldstein, and Schauser. "Active Messages: a Mechanism for Integrated Communication and Computation", 19 th International Symposium on Computer Architecture, 1992.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....via some form of shared memory must use exactly the same mechanisms as message passing applications use. However, we claim that shared memory systems have several characteristics that can be used in improving the messaging performance even further. In order to support this claim, we focus on MSAs [15, 5, 8, 1, 13, 7, 14, 6, 11, 3] that have been proposed and implemented for the Myrinet LAN and on applications running on top of pagebased software distributed shared memory systems (software DSMs) More specifically, we support our claim by studying the communication behavior of several applications running on top of the ....

T. Eicken, D. Culler, S. Goldstein, and K. Schauser. Active Messages: A Mechanism for Integrated Communication and Computation. In Proceedings of the 19th ISCA, pages 256-- 266, May 1992.

....and used. 6 Related Work Few RPC systems contain mechanisms for the parallel execution of remote software. MRPC [7] is a RPC system tuned for providing high performance for MPMD applications on homogeneous clusters. The RPC communications are imple mented on top of Active Messages (AM) [31] and the user s client programs are written in Compositional C (CC ) The work by Maassen et. al [18] extends Java RMI for efficient communications in solving high performance computing problems. Both MRPC [7] and the Java RMI extension [18] requires the end user s programs to be parallel ....

T. yon Eicken, D. E. Culler, S.C. Goldstein, K. E. Schauser, Active Messages: A Mechanism for Integrated Communication and Computation, in: 19th International Symposium on Computer Architecture, Gold Coast, Australia, 1992, pp. 256 266.

....into segments that have been bound to an endpoint that it owns. This involves matching each queued address against the list of valid pages for an endpoint. U Net specifies a software architecture for use with a U Net enabled network interface. This architecture enables use of Active Messages [20], which allow receipt of a message to result in upcalls to an application routine that is specified in the message; the Active Messages enables efficient overlapping of communication and computation in multiprocessors. U Net style ADCs allow the Active Messages concept to be extended to ....

Eicken, T. von, Culler, D.E., Goldstein, S.C., and Schauser, K.E., "Active Messages: A Mechanism for Integrated Communication and Computation," Proc. 19th ISCA, May 1992.

....communication model that bypasses the kernel from the common communication path. The VI Architecture specification [18] was developed as a joint effort by Compaq, Intel and Microsoft. It is based on previous academic research on user level communication including U Net [8] Active Messages [23] and VMMC [15] In the VI Architecture each consumer process is provided with a protected, directly accessible interface to the network hardware a Virtual Interface (VI) which represents a communication endpoint. This is different from the traditional network architecture, in which the host ....

EICKEN, T., CULLER, D., GOLDSTEIN, S., AND SCHAUSER, K. Active Messages: A Mechanism for Integrated Communication and Computation. In Proceedings of the 19th Annual Symposium on Computer Architecture (May 1992), pp. 256--266.

.... by a large processing overhead, which prevents them from fully exploiting the performance of the new networks (Figure 1) Over the past few years, many research projects have studied the design of high performance communica tion software (Fast Messages (FM) 19] Active Messages (AM) [29], U Net [28] VMMC 2 [9] PM [27] BIP [24] In the Fast Messages project, we built two generations of systems optimized to deliver communication performance to the application. The first generation, FM 1.0, was based on our studies of essen tial communication guarantees (reliable, in order ....

....communica tion in the host. However it needs not to be called for the network to make progress. FM provides buffering so that senders can make progress while their corresponding receivers are computing and not servicing the network. The FM interface is similar to the Active Messages model [29] from which it borrows the notion of mes sage handlers. However there are a number of key dif ferences: the FM API offers stronger guarantees (in particular in order delivery) uniform handling of mes sages with respect to size, and it does not follow a rigid request reply scheme. Also, in ....

[Article contains additional citation context not shown here]

T. yon Eicken, D. CulleL S. Goldstein, and K. Schauser. Active Messages: a mechanism for integrated communication and computation. In Proceedings of the International Symposium on Computer Architecture, pages 256-266, 1992.

....codes for execution. However, several recent studies on NCC machines imply that using a shared memory model with one sided communication primitives may be a better way to program an NCC machine than using a message passing model with two sided communication. For instance, some experimental studies [3, 13, 25] with micro benchmarking indicated that one sided implementations are likely more efficient on NCC machines than two sided implementations because they utilize the architectural features of the machines more efficaciously . In addition, other studies [8, 15] concluded that (1) Put Get is ....

T. Eicken, D. Culler, S. Goldstein, and K. Schauser. Active Messages: a Mechanism for Integrated Communication and Computation. International Symposium on Computer Architecture, pages 256--266, June 1992.

....that messages have arrived. Instead, they rely on applications polling frequently enough to ensure timely message handling. By default, polls of communication interfaces usually occur only when new messages are sent. This approach is even used by most implementations of Active Messages [8], whose interface suggests exactly the opposite. While polling approaches are sufficient for applications that communicate only in parallel, we argue that they perform poorly for applications that have asynchronous request response communication patterns, including most client server ....

....contribution of this paper is our finding that multi threading is an effective substitute for interrupts for our application domain. The second approach is to increase the frequency of polling in an attempt to ensure that messages are handled in a timely fashion. The standard polling procedure [6, 8] is to poll after each message send. The advantage is that these polls can be done inside the communication library, transparently to the application. This works well if messages are sent frequently (in the absence of contention) or in lockstep. However, incoming messages can wait too long if ....

[Article contains additional citation context not shown here]

T. v. Eicken, D. E. Culler, S. C. Goldstein, and K. E. Schauser, "Active Messages: a Mechanism for Integrated Communication and Computation," in Proceedings of the 19 th Annual International Symposium on Computer Architecture, May 1992.

....a test and set operation to find out if the receiving node is busy. The testand set operation requires hardware and software support for message interrupts at the receiving nodes. For the method to be efficient, the cost of this operation should be small. In this paper we use Active Messages [6] for the test and set operation. Our scheduling scheme is distributed in nature and therefore is useful even for the cases in which the same communication pattern is used only a few times (or once) In contrast, some of the algorithms we have developed [12] may be more suitable for applications ....

....and 10M bytes sec, respectively [4] The control network handles operations requiring the cooperation of many or all processors. It accelerates collective operations such as broadcast and integer reduction, and system management operations such as error reporting. 2.3. CM 5 CMAML Active Messages [6] is an asynchronous communication mechanism with the following underlying scheme: Each message header contains the address of a userlevel handler that is executed at the receiving node upon message arrival, with the message body as argument(s) The purpose of the handler is to get the message out ....

T.V. Eicken, D.E. Culler, S.C. Goldstein, and K.E. Schauser. Active Messages: a Mechanism for Integrated Communication and Computation. In Proceedings of the 19th International Symposium on Computer Architecture, Gold Coast, Australia, May 1992. ACM Press.

....and hence the levels at which improvements can be made. The lowest layer is what we will call the communication layer. This consists of the communication hardware and the low level software libraries that provide basic messaging facilities. A lot of research has been done in such libraries [7, 22, 6], and it has brought the latencies and especially bandwidths quite close to those afforded by the underlying hardware. The next layer is the protocol that provides the programming model to the parallel application programmer. We will assume all software protocols in this paper, and will focus on ....

....and is large enough that they do not constitute a bottleneck in the communication subsystem. If the network queues fill, the NI interrupts the main processor and delays it to allow queues to drain. Network links operate at processor speed and are 16 bits wide. We assume a fast messaging system [7, 21, 5] as the basic communication library. The memory bus is split transaction, 64 bits wide, with a clock cycle four times slower than the processor clock. Arbitration takes one bus cycle, and the priorities are, in decreasing order: outgoing network path of the NI, second level cache, write buffer, ....

T. Eicken, D. Culler, S. Goldstein, and K. Schauser. Active messages: A mechanism for integrated communication and computation. In Proceedings of the 19th Annual Symposium on Computer Architecture, pages 256--266, May 1992.

....i.e. it does not allow more than one independent BIP application, or even any IP traffic over the Myrinet network while the application is running. FM Fast Messages Fast Messages (FM) 8] is not just another messaging system for Myrinet network. Similar to Berkeley Active Messages (AM) [4], FM proposes a new message passing paradigm. It was developed at University of Illinois. The FM API provides services in a RISC manner, by supplying as little as possible communication primitives. The basic idea behind the FM design is to integrate communication and processing; for this, it ....

T. Eicken, D. Culler, S. Goldstein, K. Schauser, "Active Messages: A mechanism for integrated communication and computation", Proceedings of 19 th International Symposium on Computer Architecture, 1992.

....of the queues is such that they do not constitute a bottleneck in the communication subsystem. If the network queues fill, the NI interrupts the main processor and delays it to allow queues to drain. Network links operate at processor speed and are 16 bits wide. We assume a fast messaging system [5,16,4] as the basic communication library. The memory bus is split transaction, 64 bits wide, with a clock cycle four times slower than the processor clock. Arbitration takes one bus cycle, and the priorities are, in decreasing order: outgoing network path of the NI, second level cache, write buffer, ....

T. Eicken, D. Culler, S. Goldstein, and K. Schauser. Active messages: A mechanism for integrated communication and computation. In Proceedings of the 19th Annual Symposium on Computer Architecture, pages 256-266, May 1992.

....within the PE or it will be shortly moved to it (represented by the square in the picture) If the link points to another PE, then the object belongs to a tree s component. Aleph also permits the program to use the message passing approach. Messages in Aleph are loosely modeled on active messages [16] where each message encompasses a method and its arguments, which is invoked on message arrivals. New classes of messages are de ned by extending the abstract class aleph.Message. The programmer must provide then a run( method, which will be called at the receiver upon message arrival. Results ....

Eicken T, Culler D, Goldstein S, Schauser K. 1992. Active Messages: a Mechanism for Integrated Communication and Computation. In Proceedings of the 19 th Annual Int'l Symp. On Computer Architecture (ISCA'92); May 1992.

....with the same of different completion queues. For example, one completion queue could be used for all receives while the other is used for sends. 3 Active Messages Active Messages (AM) is a simple, extensible paradigm for message based communication in parallel and distributed computing systems [Eic92, Cul95]. At its core is the concept of integrating communication and computation in a way that minimizes the impact of communication overhead on overall performance. The Active Message mechanism may be viewed as essentially a specialized remote procedure call. Each message contains the name of a ....

T. von Eicken, D. E. Culler, S. C. Goldstein, K.E. Schauser. "Active messages: a Mechanism for Integrated Communication and Computation." Computer Architecture News, vol.20, (no.2), (19th Annual International Symposium on Computer Architecture, Gold Coast, Qld., Australia, May 1992. pp.256-266

....message passing. These examples show how the Library can be used to support programming models beyond the basic remote procedure call and macrodataflow program graph based styles implemented by the default library configuration. 35 5. 1 Active Messages The active messages programming model [9] is a message passing scheme that is intended to integrate communication and computation in order to increase the compute communicate overlap, Declaration of the encryption handler LegionEventHandlerStatus encryptionHandler(UVaL Reference LegionEvent ev) extract the LegionMessage from ....

....if a method (i.e. a handler) required two messages from different sources for activation, this requirement would be enforced by the Legion invocation matcher. Programs might be entirely composed of standard single token active messages, providing a programming model as flexible as the original [9]. On the other hand, programs might also include multi token active messages, for a more general programming model that might best be called active methods . 5.2 Path Expressions The various method invocation semantics covered thus far have offered a one size fits all conint ....

Thornsten von Eicken, David E. Culler, Seth C. Goldstein, and Klaus E. Schauser. Active Messages: A Mechanism for Integrated Communication and Computation. In Proceedings of the International Symposium on Computer Architecture, pages 256-266, May 1992.

....On the CM 5, an active message is a single packet message in which the first word of the message specifies a procedure on the receiving processor. This procedure is called a handler and the handler s job is to remove the message from the network and incorporate it into the ongoing computation[vCGS92]. A CM 5 active message packet consists of five words. handler arg1 arg2 arg3 arg4 When the receiving node polls the network and discovers an active message, the polling procedure invokes the handler with the four arguments arg1, arg2, arg3,andarg4 as parameters. 6.1 Sending Active Messages ....

T. von Eicken, D. E. Culler, S. C. Goldstein, and K. E. Schauser. Active messages: a mechanism for integrated communication and computation. In Proceedings of the 19 th International Symposium on Computer Architecture (ISCA '92), pages 256--266, May 1992.

....Execution Yunn Yen Chen Chung Ta King Department of Computer Science, National Tsing Hua University, Hsinchu, Taiwan 300, R.O.C. Abstract It has been shown that the large communication overhead in current multicomputer systems can be reduced greatly by using proper communication mechanisms [1] and processornetwork interfaces [4] A good processor network interface design is highly dependent on application characteristics, programming model, memory interface, and routing characteristics of the network. Unfortunately, these issues are not exploited fully in the previous work. In this ....

....through the interprocessor network. Thus, the speed and efficiency of interprocessor communication are crucial to the performance of such systems. Unfortunately, most multicomputers introduced so far have a relatively high communication to computation ratio usually two orders of magnitude [1]. It follows that these systems can only exploit coarse to medium grained computations with a limited amount of parallelism. In order to exploit massive parallelism in applications, an architecture capable of supporting fine grained computations is essential. The naive approach to improving the ....

[Article contains additional citation context not shown here]

T. von Eicken, D. E. Culler, S. C. Goldstein, and K. E. Schauser, "Active Messages: A Mechanism for Integrated Communication and Computation," in Proc. of the 19th ISCA, pp. 256 -- 267, May 1992.

....on alternative APIs becomes a key 5 requirement to achieve enhanced performance for ATMbased HPDC environments. In [17] a multicomputer communication mechanism, namely Active Messages is adapted to ATM. This mechanism is designed to implement a very fast, low granularityRPC model implementation [18]. Each transmitted message includes a pointer to a user level routine called handler whose function is to extract the message from the network and to integrate it with the ongoing computation, in a similar manner as carried out by the processor when receiving an interrupt request. Since all this ....

T. von Eicken, D. Culler, S. Goldstein, and K. Schauser. Active Messages: A Mechanism for Integrated Communication and Computation. In Proceedings of the 19th ACM International Symposium on Computer Architecture, pages 256--266, 1992.

....program. Consequently, translating all data movements into predetermined send receive pairs is not always easy. Furthermore, object oriented style lends itself more naturally to a single sided communication model 3 where a member function is invoked through a pointer to an object. Active messages [13] and remote procedure call methods must be a component of the RTS. 4. The RTS should provide a consistent interface across all the target platforms, but it should provide explicit support for hardware features that have been added to many parallel systems. A primary difference between an MPP ....

von Eicken, Thorsten; Culler, David E.; Goldstein, Seth Copen; Schauser, Klaus Erik. Active Messages: a Mechanism for Integrated Communication and Computation. UCB//CSD-92-675, March 1992. 20 pages.

....Much research has been done in the past to improve performance of both shared memory and message passing architectures. For message passing systems, researchers have largely focused on reducing the high per message overhead typically found in message passing systems. For example, active messages [von Eicken et al. 92] are a low level transport mechanism that achieves low latency by efficiently dispatching to a message handler on the receiving node. Felten [Felten 93a] proposes using a protocol compiler to custom generate message passing protocols for a given program and thus reduce protocol overhead. The ....

....for irregular distributions, all compilation approaches will incur higher overhead for address and index calculation. Note that the remote read and write operations can be implemented efficiently on a message passing communication architecture using a low level mechanism like Active Messages [von Eicken et al. 92] which provide a very low overhead data transfer mechanism, without the protocol overhead of message passing. On the other hand, we have introduced explicit synchronization operations that were performed implicitly in the message passing model. We can think of the synchronization operations as ....

[Article contains additional citation context not shown here]

T. von Eicken, D. E. Culler, S. C. Goldstein, and K. E. Schauser. Active messages: a mechanism for integrated communication and computation. In Proceedings of 19th International Symposium on Computer Architecture, pages 256--266, May 1992.

....to reach its destination; ffl o: overhead time spent by software to dequeue a message from the network interface, dispatch a message handler, and return from interrupt (or from a polling loop) ffl P : Processors number of processors in the system. LoPC assumes the Active Messages style [13] communication paradigm in which messages contain the address of their handler to facilitate quick message processing. LoPC omits LogP s g, or gap, parameter; LoPC assumes that messages may be injected into the network without interference from previously sent messages. LoPC divides computation ....

Thorsten von Eicken, David E. Culler, Seth Copen Goldstein, and Klaus Eric Schauser. Active messages: A mechanism for integrated communication and computation. In Proceedings of the 19th Annual International Symposium on Computer Architecture, Gold Coast, Australia, May 1992.

No context found.

Eicken, Culler, Goldstein, and Schauser. "Active Messages: a Mechanism for Integrated Communication and Computation", 19 th International Symposium on Computer Architecture, 1992.

No context found.

Eicken, T.v., et al. Active Messages: a mechanism for integrated communication and computation. in Proceedings of the International Symposium on Computer Architecture. 1992.

No context found.

T. Eicken, D.E. Culler, S.C. Goldstein, and K.E. Schauser. Active messages: A mechanism for integrated communication and computation. In Proceedings of the 19th International Symposium on Computer Architecture (ISCA), pages 256--266, May 1992.

No context found.

T. Eicken, D.E. Culler, S.C. Goldstein, and K.E. Schauser. Active messages: A mechanism for integrated communication and computation. In Proceedings of the 19th International Symposium on Computer Architecture (ISCA), pages 256--266, May 1992.

No context found.

T. v. Eicken et al., "Active messages: a mechanism for integrated communication and computation," in ISCA, May 1992.

First 50 documents  Next 50

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