Throughput of random arbitration for approximate matchings

Exact matchings have been extensively studied in the past, but there is relatively little work about approximate matchings. We present a throughput analysis when random arbitration is used to obtain approximate matchings. We compute the throughput by the average available credits at outputs, which is obtained through a DTMC modeling. We also devise an iterative solving procedure for the DTMC of available output credits, based on the received grants at inputs. Both our analytical and simulation results show that applying random arbitration on the studied architecture containing a moderate buffer size (B > 8) yields almost hundred percent throughput.