Ravindran and Stumm. A performance comparison of hierarchical ring- and mesh-connected multiprocessor networks. In 3rd International Symposium on High-Performance Computer Architecture. IEEE Computer Society, 1997.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....over the interconnection network that connects processing elements. In this section, performance studies in the interconnection networks for multiprocessors in the past are introduced. Ravindran and Stumm compared the performance of multiprocessors using hierarchical rings and mesh networks [61]. They used parametric simulations for their 17 study, and the miss latency was used as the performance metrics. Their study mainly showed the maximum number of nodes at which the hierarchical ring outperformed the mesh network. Since they assumed a unidirectional ring, the nearest neighbor ....

G. Ravindran and M. Stumm, "A Performance Comparison of Hierarchical Ringand Mesh-connected Multiprocessor Networks," in Proceedings of International Symposium on High Performance Computer Architecture, 58--69, February 1997.

....as in classical FPGA. Crossbar based networks are generic in term of connections but suffer from the lack of scalability. On the other hand, hierarchical ring networks which combines simple interface and high performance communications when data locality is observed have been lately studied [16]. These studies have shown the superiority of hierarchical ring networks over mesh networks. From our point of view, hierarchical ring network is well adapted to FPPA and a low power implementation of this kind of networks will be described. We first present the E FPPA architecture and the ....

....these networks and their performances are well know [15] The interconnection network which has been selected for the E FPPA is a k ary 1 cube, which are ring with k nodes. More precisely, the blocks are interconnected thanks to a hierarchical ring architecture (Figure 2a) Ravindran et al. [16] have proven that small hierarchical rings are much more efficient than mesh of higher dimension. In this architecture, each block (B) is connected to a ring of level (i) by a transfer controller (TC) which handles all the interface between the block and the ring network. Each level (i) ring is ....

G. Ravindran, M. Stumm, "A Performance Comparison of Hierarchical Ring- and Meshconnected Multiprocessor Network", in Proceedings of HPCA'97, pp. 58-69, 1997.

....in Section 4. Some conclusions are provided in Section 5. 1.1 Related Work There have been several researches in the comparisons of interconnection networks for multiprocessors in the past. Ravindran and Stumm compared the performance of multiprocessors using hierarchical ring and mesh networks [14]. The miss latency was used for performance comparison. Their study mainly showed maximum number of nodes at which the hierarchical ring outperformed the mesh network. Since they assumed a unidirectional ring, nearest neighbor communication pattern was not taken into account. Barroso and Dubois ....

....delay of a data message on the bidirectional ring is 8tr 1 . We assume the slotted ring configuration for the bidirectional ring. 3 Performance Model The methods that have been used for performance evaluation of computer systems include analytical models ( 17] parametric simulations ([14]) trace driven simulations ( 6] and execution driven simulations ( 5] In this paper, we use the hybrid approach by Barroso and Dubois [1] by extending it to the bidirectional ring and the crossbar. Below we briefly describe the derivation of execution time using the hybrid approach. 1 ....

G. Ravindran and M. Stumm, "A Performance Comparison of Hierarchical Ring- and Mesh-connected Multiprocessor Networks", in Proceedings of International Symposium on High Performance Computer Architecture, 58--69, February 1997.

....in Section 3. The comparison of the bidirectional ring and the crossbar by estimated execution times is presented in Section 4. Some conclusions are provided in Section 5. 1. 1 Related Work Ravindran and Stumm compared the performance of multiprocessors using hierarchical ring and mesh networks [4]. The miss latency was used for performance comparison. Their study mainly showed maximum number of nodes at which the hierarchical ring outperformed the mesh network. Since they assumed a unidirectional ring, nearest neighbor communication pattern was not taken into account. Barroso and Dubois ....

....multiple packets and they are sent in (possibly) uncontiguous slots. We assume the slotted ring configuration for the bidirectional ring. 3 Performance Model The methods that have been used for performance evaluation of computer systems include analytical models ( 8] parametric simulations ([4]) trace driven simulations ( 9] and execution driven simulations ( 10] In this paper, we use the hybrid approach by Barroso and Dubois [5] by extending it to the bidirectional ring and the crossbar. Below we briefly describe the derivation of execution time using the hybrid approach. The ....

G. Ravindran and M. Stumm, "A Performance Comparison of Hierarchical Ringand Mesh-connected Multiprocessor Networks ", in Proceedings of International Symposium on High Performance Computer Architecture, 58--69, February 1997.

....in Section 3. The comparison of the bidirectional ring and the crossbar by estimated execution times is presented in Section 4. Some conclusions are provided in Section 5. 1. 1 Related Work Ravindran and Stumm compared the performance of multiprocessors using hierarchical ring and mesh networks [4]. The miss latency was used for performance comparison. Their study mainly showed maximum number of nodes at which the hierarchical ring outperformed the mesh network. Since they assumed a unidirectional ring, nearest neighbor communication pattern was not taken into account. Barroso and Dubois ....

....multiple packets and they are sent in (possibly) uncontiguous slots. We assume the slotted ring configuration for the bidirectional ring. 3 Performance Model The methods that have been used for performance evaluation of computer systems include analytical models ( 8] parametric simulations ([4]) trace driven simulations ( 9] and execution driven simulations ( 10] In this paper, we use the hybrid approach by Barroso and Dubois [5] by extending it to the bidirectional ring and the crossbar. Below we briefly describe the derivation of execution time using the hybrid approach. The ....

G. Ravindran and M. Stumm, "A Performance Comparison of Hierarchical Ring- and Meshconnected Multiprocessor Networks", in Proceedings of International Symposium on High Performance Computer Architecture, 58--69, February 1997.

....Out Ring In P = Processor M = Memory NI = Network Interface I O = SCSI, Ethernet, etc. Station Bus Interface Stations NI Figure 1. NUMAchine architecture Unidirectional slotted rings were chosen for a number of reasons. First, they can perform as well as meshes for up to 128 processors [17, 18] when some data locality is present. Second, stations can be added one at a time without significant re wiring or topology changes, making them highly modular and cost effective. Third, rings exhibit two features useful for implementing cache coherence and memory consistency: inherent sequencing ....

G. Ravindran and M. Stumm. A performance comparison of hierarchical ring- and mesh-connected multiprocessor networks. In Proc. of the 3rd Intl. Symposium on High Performance Computer Architecture, pages 58--69, 1997.

.... improves throughput by increasing the buffer size at the routers to more than just a few flits, thereby reducing significantly the number of links a packet can block [59; 71] Increasing the buffer size beyond the largest worm size results in diminishing returns in network throughput, however [59]. 4.3.2 Hybrid switching. Hybrid switching [71] combines the best of both wormhole and virtual cut through switching schemes. It employs wormhole switching at lightly loaded network conditions and combines both wormhole and virtual cutthrough switching under heavy loads by selectively buffering ....

G. Ravindran and M. Stumm, "A performance comparison of hierarchical ring- and meshconnected multiprocessor networks," Proc. Intl. Symp. on High Performance Computer Architecture, pp. 58-71, Feb. 1997.

....be possible to do so such that the tasks that communicate frequently are placed close to one another so that most communication will be local to one cluster. If there is locality in the communication pattern of the applications, then hierarchical ring networks can scale to a larger number of nodes [73]. 12 Chapter 2. Issues in the Design of Multiprocessor Networks Source Destination 1 2 3 Different parts of a message in transit Established path Figure 2.5: Circuit switching in a 2 dimensional mesh network. The source and destination nodes are shown in dark and the intermediate nodes are shown ....

....from wormhole to virtual cut through at high loads by selectively buffering entire packets [85] 2. buffered wormhole switching improves throughput by increasing the buffer size at the routers to more than just a few flits, thereby reducing significantly the number of links a packet can block [73]. 7 3. wave switching combines circuit switching and wormhole switching, whereby circuit switching is used between nodes that are going to communicate frequently, while wormhole switching is used to transmit packets for which circuit switching is not efficient [24] In this dissertation we ....

G. Ravindran and M. Stumm, "A performance comparison of hierarchical ring- and meshconnected multiprocessor networks," Proc. Intl. Symp. on High Performance Computer Architecture, pp. 58-71, February 1997.

....because stations may be added to rings as necessary, and the depth of the hierarchy can be extended for more rings. Hierarchical rings, as an alternative to meshes, have been shown to perform well for systems with up to 128 processors and workloads with medium to high memory access locality [6]. A slotted ring protocol transfers data packets across the network and then reassembles these packets at the remote station. Each ring packet is a portion of a bus transaction augmented with additional routing information. The routing information is contained in a set of routing masks, one for ....

G. Ravindran and M. Stumm. A Performance Comparison of Hierarchical Ring- and Mesh-Connected Multiprocessor Networks. In Proc. of the Third International Symposium on HPCA, pages 58--69, San Antonio, Texas, February 1997.

No context found.

Ravindran and Stumm. A performance comparison of hierarchical ring- and mesh-connected multiprocessor networks. In 3rd International Symposium on High-Performance Computer Architecture. IEEE Computer Society, 1997.

No context found.

Govindan Ravindran and Michael Stumm. A Performance Comparison of Hierarchical Ring- and Mesh-Connected Multiprocessor Networks. In Proceedings of the 3rd International Symposium on High Performance Computer Architecture, pages 58--69, San Antonio, Texas, February 1997.

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