Israel Cidon

QNoC: QoS architecture and design process for network on chip

We define Quality of Service (QoS) and cost model for communications in Systems on Chip (SoC), and derive related Network on Chip (NoC) architecture and design process. SoC inter-module communication traffic is classified into four classes of service: signaling (for inter-module control signals); real-time (representing delay-constrained bit streams); RD/WR (modeling short data access) and block-transfer (handling large data bursts). Communication traffic of the target SoC is analyzed (by means of analytic calculations and simulations), and QoS requirements (delay and throughput) for each service class are derived. A customized Quality-of-Service NoC (QNoC) architecture is derived by modifying a generic network architecture. The customization process minimizes the network cost (in area and power) while maintaining the required QoS.The generic network is based on a two-dimensional planar mesh and fixed shortest path (X-Y based) multiclass wormhole routing. Once communication requirements of the target SoC are identified, the network is customized as follows: The SoC modules are placed so as to minimize spatial traffic density, unnecessary mesh links and switching nodes are removed, and bandwidth is allocated to the remaining links and switches according to their relative load so that link utilization is balanced. The result is a low cost customized QNoC for the target SoC which guarantees that QoS requirements are met.

MetaRing-a full-duplex ring with fairness and spatial reuse

The design principles of a ring network with spatial bandwidth reuse are described. A distributed fairness mechanism for this architecture, which uses low latency hardware control signals, is presented. The basic fairness mechanism can be extended for implementing multiple priority levels and integration of asynchronous with synchronous traffic. The ring is full-duplex and has two basic modes of operation: buffer insertion mode for variable-size packets and slotted mode for fixed-size packets or cells. Concurrent access and spatial reuse allow simultaneous transmissions over disjoint segments of a bidirectional ring and can increase the effective throughput by a factor of four or more. The combination of a full-duplex ring, spatial reuse, a reliable fairness mechanism, and the exploitation of advent in fiber-optic technology are the basis for the MetaRing network architecture. >

Distributed assignment algorithms for multihop packet radio networks

New distributed dynamic channel assignment algorithms for a multihop packet radio network are introduced. The algorithms ensure conflict-free transmissions by the nodes of the network. The basic idea of the algorithms is to split the shared channel into a control segment and a transmission segment. The control segment is used to avoid conflicts among nodes and to increase the utilization of the transmission segment. It is shown how these algorithms can be used in order to determine time-division multiple access (TDMA) cycles with spatial reuse of the channel. >

Analysis of multi-path routing

In connection-oriented networks, resource reservations must be made before data can be sent along a route. For short or bursty connections, a selected route must have the required resources to ensure appropriate communication with regard to desired quality-of-service (QoS). For example, in ATM networks, the route setup process considers only links with sufficient resources and reserves these resources while it advances toward the destination. The same concern for QoS routing appears in datagram networks such as the Internet, when applications with QoS requirements need to reserve resources along pinned routes. In this paper, we analyze the performance of multi-path routing algorithms and compare them to single-path reservation that might be persistent, i.e., retry after a failure. The analysis assumes that the routing process reserves resources while it advances toward the destination, thus there is a penalty associated with a reservation that cannot be used. Our analysis shows that while multi-path reservation algorithms perform comparably to single-path reservation algorithms, either persistent or not, the connection-establishment time for multi-path reservation is significantly lower. Thus, multi-path reservation becomes an attractive alternative for interactive applications such as World Wide Web browsing.

Congestion control for high speed packet switched networks

The authors suggest and investigate a general input congestion control scheme that takes into account a broad spectrum of network issues. As a preventive congestion control strategy, a leaky-bucket-type scheme operating on a session basis that limits the sessions average rate and the burstiness is proposed. This restrictive control is combined with an optimistic bandwidth usage scheme which works by marking packets into two different colors, green and red. The packets are marked so that the average green packet rate entering the network is at the reserved average rate. The average red packet rate represents traffic in excess of this guaranteed average rate and is sent to further utilize unused bandwidth in the network. Both types of packets are further filtered by a spacer which limits the peak rate at which the packets enter the network. The marked packets are then sent into the network, where they are treated according to their color, using at each intermediate node a simple threshold policy.<<ETX>>

Optimal buffer sharing

We address the problem of designing optimal buffer management policies in shared memory switches when packets already accepted in the switch can be dropped (pushed-out). Our goal is to maximize the overall throughput, or equivalently to minimize the overall loss probability in the system. For a system with two output ports, we prove that the optimal policy is of push-out with threshold type (POT). The same result holds if the optimality criterion is the weighted sum of the port loss probabilities. For this system, we also give an approximate method for the calculation of the optimal threshold, which we conjecture to be asymptotically correct. For the N-ported system, the optimal policy is not known in general, but we show that for a symmetric system (equal traffic on all ports) it consists of always accepting arrivals when the buffer is not full, and dropping one from the longest queue to accommodate the new arrival when the buffer is full. Numerical results are provided which reveal an interesting and somewhat unexpected phenomenon. While the overall improvement in loss probability of the optimal POT policy over the optimal coordinate-convex policy is not very significant, the loss probability of an individual output port remains approximately constant as the load on the other port varies and the optimal POT policy is applied, a property not shared by the optimal coordinate-convex policy. >

Bandwidth management and congestion control in plaNET

The protocols and mechanisms necessary for network bandwidth management and congestion control are addressed. The discussion draws heavily on the lessons learned from the design and implementation of plaNET, a high-speed packet-switching system for integrated voice, video, and data communications. A general overview of the mechanisms involved is given. The individual components of the system are discussed. Most of the conclusions are general and can be applied to other high-speed networks, including asynchronous transfer mode (ATM) systems.<<ETX>>

Multi-path routing combined with resource reservation

In high-speed networks it is desirable to interleave routing and resource (such as bandwidth) reservation. The PNNI standard for private ATM networks is an example of an algorithm that does this using a sequential crank-back mechanism. We suggest the implementation of resource reservation along several routes in parallel. We present an analytical model that demonstrates that when there are several routes to the destination it pays to attempt reservation along more than a single route. Following this analytic observation, we present a family of algorithms that route and reserve resources along parallel subroutes. The algorithms of the family represent different trade-offs between the speed and the quality of the established route. The presented algorithms are simulated against several legacy algorithms, including the PNNI crank-back, and exhibit higher network utilization and faster connection set-up time.

Cost considerations in network on chip

Systems on Chip (SoCs) require efficient inter-module interconnection providing for the required communications at a low cost. We analyze the generic cost in area and power of Networks on Chip (NoCs) and alternative interconnect architectures: a shared bus, a segmented bus and a point-to-point interconnect. For each architecture we derive analytical expressions for area, power dissipation and operating frequency as well as asymptotic limits of these functions. The analysis quantifies the intuitive NoC scalability advantages.Next we turn to NoC cost optimization. We explore cost tradeoffs between the number of buffers and the link speed. We use a reference architecture, termed QNoC (Quality-of-Service NoC), which is based on a grid of wormhole switches, shortest path routing and multiple QoS classes. Two traffic scenarios are considered, one dominated by short packets sensitive to queuing delays and the other dominated by large block-transfers. Our simulations show that network cost can be minimized while maintaining quality of service, by trading off buffers with links in the first scenario but not in the second.

Propagation and Leader Election in a Multihop Broadcast Environment

The paper addresses the problem of solving classic distributed algorithmic problems under the practical model of Broadcast Communication Networks. Our main result is a new Leader Election algorithm, with O(n) time complexity and O(n · lg(n)) message transmission complexity. Our distributed solution uses a special form of the propagation of information with feedback (PIF) building block tuned to the broadcast media, and a special counting and joining approach for the election procedure phase. The latter is required for achieving the linear time.