Fc3d: flow control-based distributed deadlock detection mechanism for true fully adaptive routing in wormhole networks

Abstract

Two general approaches have been proposed for deadlock handling in wormhole networks. Traditionally, deadlockavoidance strategies have been used. In this case, either routing is restricted so that there are no cyclic dependencies between channels or cyclic dependencies between channels are allowed provided that there are some escape paths to avoid deadlock. More recently, deadlock recovery strategies have begun to gain acceptance. These strategies allow the use of unrestricted fully adaptive routing, usually outperforming deadlock avoidance techniques. However, they require a deadlock detection mechanism and a deadlock recovery mechanism that is able to recover from deadlocks faster than they occur. In particular, progressive deadlock recovery techniques are very attractive because they allocate a few dedicated resources to quickly deliver deadlocked messages, instead of killing them. Unfortunately, distributed deadlock detection is usually based on crude time-outs, which detect many false deadlocks. As a consequence, messages detected as deadlocked may saturate the bandwidth offered by recovery resources, thus degrading performance. Additionally, the threshold required by the detection mechanism (the time-out) strongly depends on network load, which is not known in advance at the design stage. This limits the applicability of deadlock recovery on actual networks. In this paper, we propose a novel distributed deadlock detection mechanism that uses only local information, detects all the deadlocks, considerably reduces the probability of false deadlock detection over previously proposed techniques, and is not significantly affected by variations in message length and/or message destination distribution.

Efficient Deadlock Detection in Parallel Computer Systems with Wormhole Routing

“…This section presents simulation results of the proposed schemes and the scheme in [8] which is considered, to our knowledge, as most efficient in reducing the number of false deadlock detections. The simulations are run on 16x16 and 8x8x8 mesh and torus networks.…”

“…Although these schemes can detect all deadlocks, they may misinterpret simply-congested packets as deadlocked. A more sophisticated method to determine deadlock was proposed in [8], which, to our knowledge, performs best in detecting deadlocks accurately. It notices a sequence of blocked packets as a tree whose root is a packet that is advancing.…”

“…When the root becomes blocked later, only the packet blocked due to the root is eligible to recover. However, the accuracy of the mechanism in [8] relies on the dependency configuration of blocked packets as well as the threshold value. In general, deadlock is recovered by ejecting deadlocked packets from the network [8] or by forwarding them through a dedicated deadlock-free path [10].…”

“…Therefore, it is required that only real-deadlocked packets use the resources, as their occurrence frequency is low [11]. However, previous schemes [7,8,9] cannot distinguish between real deadlocked and blocked packets waiting longer than the given threshold. Also, they force all the packets in deadlock to recover, although it is sufficient to choose only one packet to break the deadlock.…”

“…The presence of deadlock is presumed, if a cyclic dependency among blocked packets is suspected through probe propagation, thereby reducing the number of packets detected as deadlocked considerably over previous schemes. The performance of our schemes is simulated and compared with that of a previous scheme [8], known to be most efficient in reducing the number of false deadlock detections.…”