NUGENT, S. The iPSC/2 direct-connect communications technology. In Proc. of the third conference on Hypercube concurrent computers and applications (1988), vol. 1, pp. 51--60.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....are relieved from buffer management. The price to pay is the implementation cost of flit level flow control [23] The WH switching principle origins from research on high performance interconnection networks for multiprocessor systems [7,31,36] A list of commercial machines using WH switching is [74, 75, 100, 107, 113, 124, 146]. Today, the same technique is finding its way also into LAN [12, 25, 43] and SAN [71, 154] applications. A particular form of the switching principle is standardized [73] as well, with corresponding components [108] commercially available. 2.3 Latency The packet latency is defined as the ....

NUGENT, S. The iPSC/2 direct-connect communications technology. In Proc. of the third conference on Hypercube concurrent computers and applications (1988), vol. 1, pp. 51--60.

....for one step of dense Cholesky factorization of a hypothetical frontal matrix for q = 16. It is shown in [30] that although this communication does not take place between the nearest neighbors on a subcube, the paths of all communications on any subcube are conflict free with e cube routing [40, 29] and cut through or worm hole flow control. This is a direct consequence of the fact that a circular shift is conflict free on a hypercube with e cube routing. Thus, a communication pipeline can be maintained among the processors of a subcube during the dense Cholesky factorization of frontal ....

S. F. Nugent. The iPSC/2 direct-connect communications technology. In Proceedings of the Third Conference on Hypercubes, Concurrent Computers, and Applications, pages 51--60, 1988.

....and subtransactions concepts ( 22] The logging is operational and logical to and within a block to support the data replication strategy of tables and indices. 4 FT Communication System Design The hypercube is a much used interconnection topology for parallel shared nothing multicomputers ([23], 7] Its regular and symmetric wiring enables simple routing algorithms. Two nodes communicate by sending messages. A node communicates directly with its neighbour nodes, i.e. the nodes it is connected to via the physical communication channels. A message may be relayed through intermediate ....

S.F. Nugent. The iPSC/2 Direct-Connect Communications Technology. In Proc. 3rd. Conf. on Hypercube Concurrent Computers and Applications, 1988.

....( 19] 30] The logging is operational and logical to and within a block to support the data replication strategy of tables and indexes. 4 The Design of a Fault Tolerant Communication System The hypercube is a much used interconnection topology for parallel shared nothing multicomputers ([32], 41] 10] Its regular and symmetric wiring enables simple routing algorithms. Two nodes communicate by sending messages. A node communicates directly with its neighbour nodes, i.e. the nodes it is connected to via the physical communication channels. A message may be relayed through ....

Steven F. Nugent. The iPSC/2 Direct-Connect Communications Technology. In Proc. of The Third Conference on Hypercube Concurrent Computers and Applications, volume 1, January 1988.

....network interface supports both traditional message passing and virtual memory mapped communication, it allows user programs to optimize their common cases. 6 Related Work The traditional network interface design is based on DMA data transfer. Recent examples include the NCUBE [14] iPSC 2 [13] and iPSC 860. In this scheme an application sends messages by making operating system calls to initiate DMA data transfers. The network interface initiates an incoming DMA data transfer 8 when a message arrives and interrupts the local processor when the transfer has finished so it can dispatch ....

Steven Nugent. The iPSC/2 direct-connect communication technology. In Pro- ceedings of 3rd Conference on Hypercube Concurrent Computers and Applications, pages 51--60, January 1988.

....contentions among partially reserved paths are easily resolved with centralized control. 3.1. Two way Wavelength Reservation Schemes. Connection establishment based on two way reservation under distributed control has been studied in multicomputer, telephony and high speed (non optical) networks [4, 19, 28, 44]. However, unique characteristics of WDM networks, such as the existence of many wavelength channels that may or may not be converted (or interchanged) depending on whether wavelength converters are used, make the existing distributed reservation schemes inefficient or nonapplicable to WDM ....

S. Nugent. The iPSC/2 direct-connect communications technology. In Proceedings of the 3rd Conference on Hypercube Concurrent Computers and Applications, volume 1, Jan. 1988.

....for one step of dense Cholesky factorization of a hypothetical frontal matrix for q = 16. It is shown in [36] that although this communication does not take place between the nearest neighbors on a subcube, the paths of all communications on any subcube are conflict free with e cube routing [46, 35] and cut through or worm hole flow control. This is a direct consequence of the fact that a circular shift is conflict free on a hypercube with e cube routing. Thus, a communication pipeline can be maintained among the processors of a subcube during the dense Cholesky factorization of frontal ....

S. F. Nugent. The iPSC/2 direct-connect communications technology. In Proceedings of the Third Conference on Hypercubes, Concurrent Computers, and Applications, pages 51--60, 1988.

....261 instructions on the fast path to receive and dispatch a message, plus the cost of a system call and a DMA receive interrupt. 6 Related Work The traditional method of designing network interfaces for multicomputers is based on DMA data transfer. Recent examples include the NCUBE [24] iPSC 2 [23] and iPSC 860. The network interface designs of these machines are very similar. An application sends messages by making operating system calls to initiate DMA data transfers. The network interface initiates an incoming DMA data transfer when a message arrives and interrupts the local processor ....

Steven Nugent. The iPSC/2 direct-connect communication technology. In Proceedings of 3rd Conference on Hypercube Concurrent Computers and Applications, pages 51--60, January 1988.

....cut through [9] routing capability. So, multi hop communication between two non neighbor processors is almost as fast as single hop neighbor communications if all the links between two processors are not currently used by other messages. The communication hardware uses the e cube routing algorithm [20]. In AAPC scheme, the screen is divided into P regions and kth region is assigned to processor k for k = 0; 1; P Gamma 1. Each processor simply performs P Gamma 1 communication steps exchanging pixel data according to the region assignmentwith a different processor at every step. Each ....

....steps exchanging pixel data according to the region assignmentwith a different processor at every step. Each processor must choose its communication partner at each step so that the hypercube links do not suffer congestion. A congestion free schedule for AAPC using e cube routing is given in [9, 20]. In this schedule, processor k sends its local pixel data belonging to the screen region k Phi i directly to processor k Phi i at exchange step i (for i = 1; P Gamma 1) where Phi denotes the bitwise exclusive or operation. After P Gamma 1 exchange steps, each processor gathers all ....

[Article contains additional citation context not shown here]

Nugent, S. F., The iPSC/2 direct-connect communications technology. Proceedings of Third Conference on Hypercube Concurrent Computers and Applications, Jan. 1988, pp. 51--60.

....for one step of dense Cholesky factorization of a hypothetical frontal matrix for q = 16. It is shown in [82] that although this communication does not take place between the nearest neighbors on a subcube, the paths of all communications on any subcube are conflict free with e cube routing [107, 81] and cut through or worm hole flow control. This is a direct consequence of the fact that a circular shift is conflict free on a hypercube with e cube routing. Thus, a communication pipeline can be maintained among the processors of a subcube during the dense Cholesky factorization of frontal ....

S. F. Nugent. The iPSC/2 direct-connect communications technology. In Proceedings of the Third Conference on Hypercubes, Concurrent Computers, and Applications, pages 51--60, 1988.

....for one step of dense Cholesky factorization of a hypothetical frontal matrix for q = 16. It is shown in [27] that although this communication does not take place between the nearest neighbors on a subcube, the paths of all communications on any subcube are conflict free with e cube routing [38, 26] and cut through or worm hole flow control. This is a direct consequence of the fact that a circular shift is conflict free on a hypercube with e cube routing. Thus, a communication pipeline can be maintained among the processors of a subcube during the dense Cholesky factorization of frontal ....

S. F. Nugent. The iPSC/2 direct-connect communications technology. In Proceedings of the Third Conference on Hypercubes, Concurrent Computers, and Applications, pages 51--60, 1988.

....matching following a single all to all personalized communication. Such a communication would allow selection of matching edges that are shared between any two processors. In a hypercube with cut though routing, it is possible to pick congestion free paths for all to all personalized communication [30, 32]. If such an all to all personalized communication scheme is used then the dominant cost of reducing G i (P ) to G i 1 (P ) would change to (c 1 t w ) N i =P ) decreasing the total cost of shrink to O Gamma N P Delta . However, this would still not change the overall complexity of SSND. In ....

S. F. Nugent, The iPSC/2 direct-connect communication technology, in Third Conference on Hypercubes, Concurrent Computers, and Applications, 1988, pp. 51--59.

....operation. Since these machines involve the operating system to handle messages, the latency of sending messages can be quite high. A simple send with small messages takes 267 s on the iPSC 2 and 437 s on an NCUBE four [Bra88] Much of this time is due to high overhead operating system routines [Nug88] One group of users rewrote parts of the nCUBE 2 operating system and reduced the overhead by nearly an order of magnitude[vCGS92] However, even the remaining time was still quite large (11 15 s) due to the expense of the DMA instructions and instructions which switch to and from the operating ....

Steven F. Nugent. The iPSC/2 direct-connect communications technology. In Proc. of the Third Conf. on Hypercube Concurrent Computers and Applications, pages 51--60, January 1988.

....instructions on the fast path to receive and dispatch a message, plus the cost of a system call and a DMA receive interrupt. 6 Related Work The traditional method of designing network interfaces for multicomputers is based on DMA data transfer. Examples include the NCUBE [22] iPSC 2 and iPSC 860 [21]. The network interface designs of these machines are very similar. An application sends messages by making operating system calls to initiate DMA data transfers. The network interface initiates an incoming DMA data transfer when a message arrives and interrupts the local processor when the ....

Steven Nugent. The iPSC/2 direct-connect communication technology. In Proceedings of 3rd Conference on Hypercube Concurrent Computers and Applications, pages 51--60, January 1988.

....Since these machines involve the operating system to handle messages, the latency of sending messages can be quite high. A simple send with small messages takes 267 s on the iPSC 2 and 437 s on an NCUBE four [Bra88] Much of this time is due to high overhead operating system routines [Nug88] One group of users rewrote parts of the nCUBE 2 operating system and reduced the overhead by nearly an order of magnitude[vECGS92] However, even the remaining time was still quite large (11 15 s) due to the expense of the DMA instructions and instructions which switch to and from the ....

Steven F. Nugent. The iPSC/2 Direct-Connect Communications Technology. In Proc. of the Third Conf. on Hypercube Concurrent Computers and Applications, pages 51--60, January 1988.

....while the torus algorithm requires two to break cycles within a single 16 dimension. These algorithms do not provide any adaptivity, but are simple and fast. There are many machines that use oblivious routing in their networks. Machines with oblivious hypercube networks include the Intel IPSC 2 [Nug88] oblivious mesh networks are found in the Intel Delta [Cor91] the Intel Paragon, the Alewife [ABC 95] the J Machine [ND92] the M Machine [FKD95] the Ametek 2010 [SAF88] and the SHRIMP [BLA 94] computers, while the Cray T3D [ST94] uses an oblivious torus. S D 1 0 Figure 2.2: An ....

....that might otherwise be implemented in these other layers. Nevertheless, nearly all known routers support a single class of routing, and few provide services such as in order delivery required by many applications. 4. 2 Motivation Oblivious routing continues to be the routing algorithm of choice [Nug88, Cor91, ABC 95, ND92, FKD95, SAF88, BLA 94, ST94] Oblivious routers are simple and provide in order delivery of messages. Since oblivious routers have no adaptivity, they perform poorly in congested networks and fail when faults are present in the network. Minimal adaptive routers are ....

S.F. Nugent. The iPSC/2 direct-connect communications technology. In Conf. on Hypercube Concurrent Computers and Applic., pages 51--60, 1988.

....In iPSC 2 hypercube multicomputer, with DCM technology, communication between two non neighboring processors is almost as fast as neighbor communications if all the links between two processors is not currently used by other messages. The communication hardware uses the e cube routing algorithm [9]. Using DCMs, we can exchange messages between non neighbor processors by the following algorithm [10] This algorithm ensures that at each exchange step the pixel data is directed to destination processors with the pixel data following disjoint paths. m : this node s id B k : the pixel data ....

S.F. Nugent, The iPSC/2 direct-connect communication technology. In Proc. Third Conf. Hypercube Concurrent Comput. and Appl., 51-60 (Jan. 1988).

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