Characterization of deadlocks in k-ary n-cube networks

Characterization of Deadlocks in Irregular Networks

“…Previous work [14,15] further establishes that deadlocks can be characterized by three attributes: deadlock set, resource set, and knot cycle density. The deadlock set is the set of messages that own the virtual channels involved in the knot and represents the expanse of the deadlock in terms of the messages involved.…”

Efficient Deadlock Detection in Parallel Computer Systems with Wormhole Routing

“…However, the frequency of deadlock occurrence is reported to be very low with a fully adaptive routing algorithm [11]. Hence, it is believed that a complex cycle would rarely occur except in a heavy network condition.…”

“…To depict resource dependencies at a point of time, the channel wait-for graph (CWFG) can be used, where vertices represent the resources (either virtual channels or physical channels for networks with no virtual channel) and edges represent either 'wait-for' or 'owned-after' relations [5,11,12]. A wait-for edge (c i , c j ) represents that there exists a message occupying channel c i and waiting for channel c j .…”

“…The frequency of deadlock occurrence is reported to be very low with a fully adaptive routing algorithm [11]. Hence, it is wasteful to limit routing adaptivity for rarely occurring deadlocks.…”

“…In heavily loaded networks, those packets presumed as deadlocked will saturate the recovery resources, thus degrading performance considerably. 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.…”