Distributed stream processing systems (DSPSs) have many important applications such as sensor data analysis, network security, and business intelligence. Failure management is essential for DSPSs that often require highly-available system operations. In this paper, we explore a new predictive failure management approach that employs online failure prediction to achieve more efficient failure management than previous reactive or proactive failure management approaches. We employ light-weight stream-based classification methods to perform online failure forecast. Based on the prediction results, the system can take differentiated failure preventions on abnormal components only. Our failure prediction model is tunable, which can achieve a desired tradeoff between failure penalty reduction and prevention cost based on a user-defined reward function. To achieve low-overhead online learning, we propose adaptive data stream sampling schemes to adaptively adjust measurement sampling rates based on the states of monitored components, and maintain a limited size of historical training data using reservoir sampling. We have implemented an initial prototype of the predictive failure management framework within the IBM System S distributed stream processing system. Experiment results show that our system can achieve more efficient failure management than conventional reactive and proactive approaches, while imposing low overhead to the DSPS.

In this paper, we present a new online failure forecast system to achieve predictive failure management for fault-tolerant data stream processing. Different from previous reactive or proactive approaches, predictive failure management employs failure forecast to perform informed and just-in-time preventive actions on abnormal components only. We employ stream-based online learning methods to continuously classify runtime operator state into normal, alert, or failure, based on collected feature streams. We have implemented the online failure forecast system as part of the IBM System S stream processing system. Our experiments show that the on-line failure forecast system can achieve good prediction accuracy for a range of stream processing software failures, while imposing low overhead to the stream system.

In this paper, we formulate a timing-driven obstacle-avoiding X-architecture Steiner minimal tree (TOA-XSMT) problem, and propose a timing-driven routing tree construction which simultaneously minimizes the maximum source-to-terminal delay and the total wirelength. First, we construct a spanning graph by the terminals and the obstacles. Second, a minimal spanning tree is obtained in a spanning graph. Third, we transform a spanning tree into a feasible X-architecture tree. Fourth, for each terminal of the routing tree, the delay is computed by a modified Elmore delay model. Fifth, an efficient rerouting method is used to improve all timing violations which their delay are over a user-defined threshold. Finally, the critical terminals are rerouted by splitting and merging procedure. Compared to the result without rerouting, the maximum source-to-terminal delay is reduced by 60.8% with the 0.7% additional total wirelength.

A pipelined common buffer architecture (PCBA) is proposed for memory based switching systems. Unlike previously proposed shared-memory switching system, the PCBA is suitable for prioritized and multicast (without multiple memory write) traffic for both fixed length and variable length cell/packet. Therefore, switching among ATM traffic and existing LAN can be accomplished. The PCBA can route both fixed-sized cell data and variable-sized packet data. It separates switching system control and packet/cell data streams (the dichotomy) inside the switching system to simplify the design process for identifying VLSI chips implementation. The PCBA also uses pipelining for both system control and data movement, to achieve the highest possible system throughput, i.e. one system buffer per system clock cycle. A system buffer in PCBA has fixed size N/spl times/W bits where N is the number of switch ports and W is the width of data path (number of bits transmitted/received to/from system buffer at one clock cycle by a switch port). Furthermore, the usage of external memory modules for both data and control memory takes full advantage of advances in commercial RAM technology. It can be seen that the PCBA not only can easily expand its queue size for different networking environments, but can also support future traffic types without modification to the architecture.>

Multiple-channel architectures have been proposed for broadcasting-based local area networks to overcome the problem of deteriorating performance when a very high speed communication channel is used. Such networks are called multiple-channel local area networks (M-LANs). In a carrier-sense multiple-access with collision detection (CSMA/CD) protocol designed for M-LANs (M-CSMA/CD), a ready station selects only one channel to transmit its packet according to some channel selection mechanism. A new CSMA/CD protocol for M-LAN architecture is proposed by which a station may select and try to transmit the same packet on K channels (K>or=1). If it has been successfully transmitted on more than one channel after the end-to-end propagation delay, it randomly selects one channel to finish its transmission and immediately stops its transmission on the rest of the channels. This is called the multipacket multichannel CSMA/CD protocol (MM-CSMA/CD). The performance of the proposed MM-CSMA/CD protocol can be shown to be potentially better than that of the M-CSMA/CD protocol and the traditional CSMA/CD protocol.>

The twin-bus architecture for voice/data integration is discussed and a hybrid protocol for the data bus that combines CSMA/CD (carrier sense multiple access with collision detection) and announcement reservation schemes is proposed. To improve the performance on data bus, part of the bandwidth on the voice bus is used as the control subchannel for announcement reservation on the data bus. Simulation results show that the proposed hybrid protocol outperforms both CSMA/CD and the announcement reservation scheme in all cases. Since announcement reservation has good performance when traffic load is heavy, the combined protocol becomes efficient for both light and heavy loads.>