Software-based deadlock recovery technique for true fully adaptive routing in wormhole networks

Abstract

Abstract

Characterization of Deadlocks in Irregular Networks

“…If deadlock is detected, it is immediately ''broken'' by removing a single randomly selected message in the deadlock set from the network in a flit-by-flit manner so as to emulate a recovery procedure similar to that described in [18]. Deadlock frequency is measured in relative terms by the normalized number of deadlocks -which is the ratio of the number of deadlocks to the number of messages delivered.…”

Performance Tuning of Adaptive Wormhole Routing through Selection Function Choice

“…For meshes, opt-y has the fewest restrictions of any fully adaptive routing algorithm that relies on routing restrictions to avoid deadlock. (Other techniques for avoiding deadlock exist, including deadlock resolution [17] and misrouting [15].) By using opt-y in the majority of the simulations, we increase the flexibility of the selection function and enhance the opportunity for improving performance.…”

Performance Evaluation of Probe-Send Fault-tolerant Network-on-chip Router

“…The default value of router parameters used are flit size equal to phit size of 512 bits (L1 cache line size of Itanium 2 processor [5]), 4 VC [3], deadlock detection timer value of 256 cycles ( [6] suggests it to be good number for avoiding false deadlock detection), and deadlock buffer size of 1024 entries (sufficient for explored network sizes). Network size is currently limited to 49 nodes due to Cadence VHDL simulation environment constraints.…”