分散处理论2.ppt

分散処理論２ (No8) Fault-tolerant routing Performance and fault tolerance are two dominant issues for interconnection networks. Techniques used to realize fault tolerance are often at the expense of considerable performance degradation. Even the failure of a single link can destroy the deadlock freedom properties. misrouting Misrouting can avoid deadlocks caused by faulty components. It must be controlled so that livelock is avoided and newly introduced dependencies do not produce deadlock. Fault-recovery mechanisms must recover messages indefinitely blocked on faulty components. Channel and network redundancy A network is said to be connected with respect to a rouging algorithm if the routing function can route a message between any pair of nonfaulty routing nodes. A channel is said to be redundant iff, after removing it, the resulting routing function is still connected and deadlock-free. A routing function is said to be f fault-torelant if for any f failed components in the network, the routing function is still connected and deadlock-free. Redundancy level A routing function has a redundancy level equal to r iff after removing any set if r channels, the routing function remains connected and deadlock-free, and there exists a set of r+1 channels such that, after removing them, the routing function is no longer connected or it is not deadlock-free. Fault regions Fault model The patterns of component failures and expectations about the behavior of processors and routers in the presence of these failures determines the approaches to achieve deadlock and livelock freedom. On a node (PE or router) failure, all physical channels incident on the failed node are also marked faulty at adjacent routers. On a link failure, all VCs on that particular physical link are marked faulty. Fault-tolerant model attributes An example of misrouting Chaotic routing Incorporates randomization to produce a nonminimal fault-tolerant routing algorithm using VCT switching. Unsafe/faulty nodes Gra