RapidlO Traffic Management and Flow Arbitration Protocol.

The RapidlO data-stréaming logical layer provides a segmentation and reassembly service to higher layers. Since a RapidlO receiver can potentially receive 64-kbyte PDUs from 64-kbyte senders, the memory requirements for successfully reassembling fragments is huge (∼4 Gbytes). As RapidlO-based hardware chips have limited memory, efficient arbitration algorithms/protocols are needed that can fairly share memory among competing traffic streams, under constraints imposed by the standard. In this article we show that absence of proper arbitration can lead to deadlocks and system underutilization. We then develop a Flow Arbitration Protocol (now a part of RapidlO 2.0) as an extension to existing specification. A defining characteristic of this protocol is that memory resources can be reserved for single- and multi-PDU transfers. [ABSTRACT FROM AUTHOR]