Adaptive Reliable Multi ast

Jaehee Yoon Azer Bestavros Ibrahim Matta jaeheey s.bu.edu bestavros s.bu.edu matta s.bu.edu Computer S ien e Department Boston University Boston, MA 02215 September 1999 Abstra t An in reasing number of appli ations, su h as distributed intera tive simulation, live au tions, distributed games and ollaborative systems, require the network to provide a reliable multi ast servi e. This servi e enables one sender to reliably transmit data to multiple re eivers. Reliability is traditionally a hieved by having re eivers send negative a knowledgments (NACKs) to request from the sender the retransmission of lost (or missing) data pa kets. However, this Automati Repeat reQuest (ARQ) approa h results in the well-known NACK implosion problem at the sender. Many reliable multi ast proto ols have been re ently proposed to redu e NACK implosion. But, the message overhead due to NACK requests remains signi ant. Another approa h, based on Forward Error Corre tion (FEC), requires the sender to en ode additional redundant information so that a re eiver an independently re over from losses. However, due to the la k of feedba k from re eivers, it is impossible for the sender to determine how mu h redundan y is needed. In this paper, we propose a new reliable multi ast proto ol, alled ARM for Adaptive Reliable Multi ast. Our proto ol integrates ARQ and FEC te hniques. The obje tives of ARM are (1) redu e the message overhead due to NACK requests, (2) redu e the amount of data transmission, and (3) redu e the time it takes for all re eivers to re eive the data inta t (without loss). During data transmission, the sender periodi ally informs the re eivers of the number of pa kets that are yet to be transmitted. Based on this information, ea h re eiver predi ts whether this amount is enough to re over its losses. Only if it is not enough, that the re eiver requests the sender to en ode additional redundant pa kets. Using ns simulations, we show the superiority of our hybrid ARQ-FEC proto ol over the well-known S alable Reliable Multi ast (SRM) proto ol.