A. Spector. Performing Remote Operations Efficiently on a Local Computer Network. Communications of the ACM, pages 246--260, April 1982.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....registers and buffer queues in the descriptor. This enables RDMA, which allows the network adapter to reduce copy overhead by accessing application buffers directly. The combination of user level network access and copy avoidance has a lengthy heritage in research systems spanning two decades [2, 4, 6, 33, 38]. The experiments in Section 7 quantify the improvement in access overhead that DAFS gains from RDMA and transport offload on direct access NICs. 2.3 User Level File Systems In addition to overhead reduction, the DAFS protocol leverages user level networking to enable the network file system ....

A. Spector. Performing Remote Operations Effi- ciently on a Local Computer Network. Communications of the ACM, 25(), pages 246-260, April 1982.

....static software DSM approach appears to remain fairly limited. Dynamic Approaches Dynamic software DSM systems typically support a more general programming model than their static counterparts, typically allowing multiple independent threads of control to operate within the shared address space [4, 5, 11, 21, 37, 52, 75]. Given mechanisms for inter thread synchronization (e.g. semaphores, barriers) a programmer is able to express essentially any form of parallelism. For the most part, these systems utilize a data shipping paradigm in which threads of computation are relatively immobile and data items (or ....

Alfred Z. Spector. Performing Remote Operations Efficiently on a Local Computer Network. Communications of the ACM, pages 246--260, April 1982.

....state in an NSM. sent common data areas. Communication between processes, then becomes an artifact of sharing some subset of one process s state space with another process s state space. Network Shared memory fits elegantly into this paradigm. Over the past decade [Smith Maguire 89] Spector 82] it has been observed that a page fault, traditionally serviced by a local disk device, could be serviced remotely. In the simplest case, this might consist of loading pages from a remote disk using a network. A more interesting case is the sharing of memory; shared pages are implemented with ....

A. Z. Spector. Performing Remote Operations Efficiently on a Local Area Network. Communications of the ACM 25(4), pp. 246-260, April 1982.

....aid in dealing with these complexities, several attempts have been made over the last decade to extend object oriented semantics to distributed systems. Examples include Eden [Lazowska 81] and Emerald [Black 86] The development of efficient mechanisms for remote procedure call [Nelson 81, Spector 82] has made distributed computing an attractive alternative to timesharing on large, general purpose mainframes. A parallel development in distributed system design was fault tolerance. Several distributed, object oriented systems have been developed that provide mechanisms for the construction of ....

....execute (as there might be in a tightly coupled multiprocessor system) the exact sequence of process interactions is unpredictable. Since the semantics of Smalltalk message passing imply a blocking send, Distributed Smalltalk follows a variation of the traditional remote operation model [Spector 82] the sending process blocks, the receiving process performs the operation, a value is returned, and the sending process then resumes. The variations have to do with the point in time 68 at which the receiving process is created and whether or not values are returned. In Distributed Smalltalk ....

Alfred Z. Spector. Performing Remote Operations Efficiently on a Local Computer Network. CACM, 25(4):246--260, April 1982.

....implementations of the remote access model on loosely coupled systems, i.e. without specific support from the network adapter. There were able to demonstrate a performance improvement over traditional message passing, mainly as a result of reduced context switch overhead on the receiving node [23, 27, 28]. It is possible to implement remote memory accesses more efficiently, i.e. reduce host overhead, by using a more powerful network adapter. A first approach is to rely on a powerful outboard processor to take over some of the communication tasks normally performed by the host, e.g. similar to the ....

.... updates, or updates under software control (i.e. the application issues a PUT) The deposit model on iWarp implements this model entirely in software [25] Finally in the right most column, a number of projects have explored the support of a remote PUT and GET for networks connected by LANs [23] [28] 27] These systems typically rely on remote procedure calls to transfer commands and data kernel tokernel. They can be more efficient than traditional send receive operations because the application does not have to be scheduled (i.e. data transfer without control transfer) Clearly, given ....

Alfred Z. Spector. Performing Remote Operations Efficiently on a Local Computer Network. Communications of the ACM, 25(4):246--260, April 1982.

....requires the application program to actively check for incoming packets. Another common idea in data movement packages is supporting read and write operations on remote processors memory. This was originally proposed by Spector, who provided a microcoded implementation on networked workstations [Spector 82] A later version of the same idea appears as part of the Active Messages system. Thekkath [Thekkath et al. 93] has proposed a general remote memory operation system suitable for multicomputers or for networked workstations. The forthcoming Cray T3D system supports a NUMA model, in which each ....

....several classes of related work to consider. 8.1.1 Data Movement Research on data movement has shown that low level data transport can be provided efficiently in a variety of ways. Spector used custom microcode to implement fast remote access instructions on 1980 vintage networked workstations [Spector 82] Thekkath updated this idea by re implementing it on today s fast workstations and networks [Thekkath et al. 93] Delp and co workers devised MemNet, which provides a set of workstations with a region of physical memory that is kept coherent by transmitting all writes around a ring ....

A. Z. Spector. Performing remote operations efficiently on a local computer network. Communications of the ACM, 25(4):246--260, April 1982. 136

....By contrast, network software performance has not risen to meet the challenge implied by the new applications and network hardware. Network software latency is especially disappointing it was high compared to network hardware latency even shortly after the invention of the Ethernet BibRef[60], and has declined comparatively slowly since. Seven years have passed between the introduction of the Sun 3 60 (1987) and the Alpha 3000 800 (1994) The Alpha is roughly 40 times faster than a Sun 3 60, implying that CPU speeds have improved by an average factor of 1.7 each year, yet the ....

....a TCP fastpath, while BibRef[51] describes a UDP fastpath. There has been more investigation into reduction of and minimization of latency for simple RPC protocols. Impressive speedups were achieved by implementing some RPC functionality in the microcode of an early Ethernet interface BibRef[60]. More recently, Ethernet interfaces have become standardized to the extent that the functionality previously implemented by a processor is implemented in cheap MAC chips there is no microcode to change. However, FDDI and ATM are new enough that adapters for them still frequently include onboard ....

[Article contains additional citation context not shown here]

A. Z. Spector, "Performing Remote Operations Efficiently on a Local Computer Network," Communications of the ACM, pp. 246-260, April 1982.

....data sharing directly. Data sharing is still possible with message passing, by maintaining the shared data in a dedicated process and operating on the data by sending operators to this process[26] Other methods may involve moving data around explicitly using message passing primitives. Spector[36], for example, observed that remote references might offer a better model for efficient communications. Remote procedure call (RPC) 5] was introduced to provide a procedure call like communications interface. Since the procedure call is executed in a separate address space, it is difficult for ....

....basically those of shared memory. The major limitations are imposed by implementation constraints (e.g. limited copying of complex data structures) 2 1. 2 Latency and Shared Address Spaces A shared memory space supports data sharing with very little overhead (and hence communication) 17] 11][36] However, there are technical difficulties in providing complete shared memory semantics in a decentralized setting. A major impediment to high performance is the latency of remote references. Latency is the time required for a memory reference and can be represented as a ratio of remote ....

[Article contains additional citation context not shown here]

Alfred Z. Spector. Performing remote operations efficiently on a local area network. Communications of the ACM, April 1982.

....on the object. Afterwards the server sends a reply message back to the client, which unblocks the client. We have named this request reply exchange a transaction (not to be confused with data base transactions) Amoeba guarantees at most once execution of transactions. Remote procedure calls [13 14] are implemented by marshalling an opcode and its arguments in a request message, and performing a transaction with the appropriate server. The result of the procedure is retrieved from the reply message. After starting a transaction, a client process blocks to await the reply. All objects in ....

Spector, A.Z.: "Performing Remote Operations Efficiently on a Local Computer Network, " Commun. ACM, vol. 25, no. 4, pp. 246-260, Apr. 1982.

....they do not address the issue of software support for running sequential, off the shelf applications on a scalable resolution display wall. The idea of executing graphics primitives remotely can be found in X windows. The mechanisms for remote procedure call [15, 2] remote execution model [19] have been investigated in the context of programming languages. The tools such as VDD and GRL described in this paper leverage and extend these ideas for remote graphics primitive executions for scalable resolution displays. VDD intercepts primitives in a device driver and executes them remotely ....

Alfred Z. Spector. Performing remote operations efficiently on a local computer network. Communications of the ACM, 25(4):260--273, April 1982.

....it must be ensured that the agent is never lost and hence will eventually get its job done. Moreover, failures may not cause the agent to perform operations more than once (e.g. to reserve and pay two seats instead of one) The exactly once property has been already defined for RPC systems [Spe82], where it defines the failure semantics of a single remote procedure. In the context of mobile agents, a sequence of agent stages are to be considered rather than a single procedure. An agent execution is defined to be exactly once if the entire sequence of its stages is eventually performed, ....

Spector, A.: "Performing remote operations efficiently on a local computer network", Communications ACM, vol. 25, pp 246-260, Apr. 1982

....and performance) current commodity networks such as Tandem s ServerNet [40] and Myrinet [13] At the network interface, SHRIMP uses its automatic and deliberate update mechanisms to support particular parallel programming models and constructs. This work relates to several prior efforts. Spector [39] proposed a remote memory reference model to perform communication over a local area network and the implementation is programmed in a processor s microcode. This model has been revived by Thekkath et al. 41] using fast traps. Druschel et al. 19] proposed the concept of application device ....

Alfred Z. Spector. Performing Remote Operations Efficiently on a Local Computer Network. Communications of the ACM, 25(4):260--273, April 1982.

....approach to support shared memory. These systems do not provide a mechanism for high bandwidth, lowoverhead block data transfer such as deliberate update. Some systems provide communication via direct access to remote memory locations. An early example is Spector s work on a network of Xerox Altos [30]. The most recent example of this approach is the CRAY T3D [6] The disadvantage of this approach is that shared data is stored in only one place, so all but one of the sharers must access it remotely. By 10 contrast, our automatic update scheme lets a producer and consumer share data without ....

Alfred Z. Spector. Performing remote operations efficiently on a local computer network. Communications of the ACM, 25(4):246--260, April 1982.

.... operations in the presence of different failure conditions, such as loss of messages or the crashes of the client or the server processes, first appeared in reference [4] A system supporting remote operations in a local area network with different reliability semantics is described in reference [21]. In most of the these RPC systems the design goals, in general, have been reliability[3] 4] 20] 22] performance[2] 21] 23] 24] and support for heterogeneous systems [19] A critique of the RPC paradigm is presented in reference [25] One of the criticisms of this paradigm is that it is ....

.... processes, first appeared in reference [4] A system supporting remote operations in a local area network with different reliability semantics is described in reference [21] In most of the these RPC systems the design goals, in general, have been reliability[3] 4] 20] 22] performance[2] [21] [23] 24] and support for heterogeneous systems [19] A critique of the RPC paradigm is presented in reference [25] One of the criticisms of this paradigm is that it is inherently sequential in nature and does not allow the flexibility of exploiting parallelism found in many distributed ....

A.Z. Spector, `Performing Remote Operations Efficiently on a Local Computer Network', Communications of the ACM, 246--259 (1982).

....model where copies are performed by both sender and receiver; and two times higher if only one copy takes place. Total CPU overhead is four times lower in SHRIMP II deliberate update than in send receive communication. 7 Related work Spector proposed a remote reference remote operation model [9] in which a master process on a local processor performs remote reference and a slave process on another remote processor implements this reference by executing the remote operation. Compared to this model, VMMC is much simpler, as we envision hardware support only for data transfer. We also use ....

A. Z. Spector. Performing remote operations efficiently on a local computer network. Communications of the ACM, 25(4):260--273, April 1982.

....are organized hierarchically. The following two sections describe these two hierarchies of channels. 3.1.1 Point to point Channels A point to point channel is characterized by its semantics. Typically, a channel can enforce the maybe, at least once, at most once, or exactly once semantics [5]. A point to point channel is also characterized by the type of the synchrony that it enforces. We distinguish three kinds of synchrony: request reply (upon a call, the client is blocked until the receipt of the reply) synchronous (upon a call, the client is blocked until the invocation message ....

Alfred Z. Spector. Performing remote operations efficiently on a local computer network. ACM, 25(4):246-- 260, April 1982.

....on the object. Afterwards the server sends a reply message back to the client, which unblocks the client. We have named this request reply exchange a transaction (not to be confused with data base transactions) Amoeba guarantees at most once execution of transactions. Remote procedure calls [10, 11] are implemented by assembling an operation code and its arguments in a request message, and performing a transaction with the appropriate server. The result of the procedure is retrieved from the reply message. After starting a transaction, a client process blocks to await the reply. A server ....

Spector, A. Z., "Performing Remote Operations Efficiently on a Local Computer Network," Comm. ACM , Vol. 25, No. 4, pp. 246-260, April 1982. -

....a runtime system that is already in widespread use, we were able to run several realistic applications and understand the impact of our hardware and software in the light of this experience. An alternative we considered was to use more specialized communication primitives, such as Remote Memory [20], Active Messages [27] etc. While offering good performance, this alternative has the disadvantage that we would have to rewrite many applications to use these primitives. 1.1. Related work The idea of running parallel programs on workstation clusters has existed for many years. A few of the ....

....good performance, this alternative has the disadvantage that we would have to rewrite many applications to use these primitives. 1.1. Related work The idea of running parallel programs on workstation clusters has existed for many years. A few of the early examples are Spector s work on the Alto [20], the Multisatellite star work in the V system [11] and the Nectar project at CMU [3] There have also been several software systems that support a shared memory or a shared object abstraction on workstation clusters for executing parallel programs. Some examples are IVY [16] Amber [10] Munin ....

A. Z. Spector. Performing remote operations efficiently on a local computer network. Commun. ACM, 25(4):246--260, April 1982.

No context found.

A. Spector. Performing Remote Operations Efficiently on a Local Computer Network. Communications of the ACM, pages 246--260, April 1982.

No context found.

A. Spector. Performing Remote Operations Efficiently on a Local Computer Network. Communications of the ACM, pages 246--260, April 1982.

No context found.

A.Z.Spector `Performing remote operations efficiently on a local computer network' Commun. ACM 25, 4 (April 1982) 246-260.

No context found.

A. Spector, Performing remote operations efficiently on a local computer network, Communications ACM 25 (1982) 246-260.

No context found.

Alfred Z. Spector. Performing remote operations efficiently on a local computer network. ACM, 25(4):246--260, April 1982.

No context found.

A.Z.Spector `Performing remote operations efficiently on a local computer network' Commun. ACM 25, 4 (April 1982) 246-260.

No context found.

Spector, A.Z. "Performing Remote Operations Efficiently on a Local Computer Network," Commun. ACM, vol. 25, pp. 246-260, April 1982.

First 50 documents

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