Fault Aware Dynamic Adaptive Routing Using LBDR - VLSI Design and Test - page 310

Ta b l e 2 . Average latency(per flit) at high congestion level

Number of Failed Links DyAD-Basic FADAR-LBDR

Failures Average Latency Average Latency

0 No link fail 10.4444 10.4444

1 5-9 infinite 10.4444

2 5-9,9-10 infinite 10.4444

3 5-9,9-10,5-6 infinite 10.4444

4 5-9,9-10,5-6,1-5 infinite 10.4444

5 5-9,9-10,5-6,1-5,6-10 infinite 24.4444

5.3 Experiment-3

In this case node 1 and 10 is chosen as source and destination nodes respectively. This
experiment is performed to show the case when basic DyAD scheme gives a single
path for few source destination pair and when that path becomes faulty. As shown in
Figure 3, there is only a single path for nodes present in even column destined for nodes
present in next odd column. Table 2 shows the results when single path becomes faulty.
As shown in Table 2, basic DyAD fails to generate any alternative path as it uses odd-
even routing. On the other side, DyAD-LBDR uses deterministic mode where it uses
XY routing to route the traffic through alternative path.

6 Result Analysis and Future Work

In this paper we have presented a Fault Aware Dynamic Adaptive Routing using LBDR
(FADAR-LBDR) in NoC. The proposed method is implemented with LBDR without
using routing table. Experimental results have shown the effectiveness of the proposed
technique. Proposed method uses minimal path till there is at least one minimal path
from source to destination. When there is no minimal path available, it uses deroute bit
for non-minimal path selection. Results shown in Table 1 demonstrates that average la-
tency and packet delivery ratio remains same in the presence of single and multiple link
failures when atleast one minimal path is avaiable. When all minimal paths encounter
faulty components, non-minimal path selection is used which results in increased la-
tency as shown in last row of Table 1. 100% packet delivery ratio is observed in case
of multiple source destination pair and multiple link failures with some increased aver-
age latency. Comparison of proposed method with basic DyAD scheme has shown in
Table 2. Proposed method keeps the average latency same for cases where basic DyAD
generates single path from source to destination. Proposed method is effective in sense
that fault tolerance is achieved without using routing tables while maintaining the same
performance. Deadlock freedom has been achieved using two virtual channels with lim-
iting cases for transition. In future, we will be addressing the non-restriction switching
of routing algorithm.

Acknowledgment. We would like to acknowledge support received from Indo-Spain
DST project DST / INT / S pain / P 35 / 11 / 1. Authors would like to acknowledge the

critical comments and inputs received from Dr. Jose Flich, DISCA, UPV, Spain for the
above work.

−