DiCoR: Distributed Cooperative Repair of Multimedia Broadcast Losses Saqib Raza Dept. of Computer Science University of California, Davis Gene Cheung Hewlett-Packard Laboratories Tokyo, Japan Chen-Nee Chuah Dept. of Electrical & Computer Engineering University of California, Davis Abstract—Multimedia Broadcast/Multicast Service (MBMS) allows a common broadcast channel to be shared by users inter- ested in identical content. We explore the problem of enhancing MBMS resilience by repairing packets lost during broadcast. Since MBMS broadcast consumes expensive 3G resources, we leverage the ubiquity of multi-homed mobile devices i.e., devices having both cellular and IEEE 802.11 wireless interfaces. We thus accomplish out-of-band repair of MBMS packet losses through an ad-hoc, peer-to-peer 802.11-based network. A fundamental challenge in scheduling repair transmissions is handling inter- ference between distributed nodes. We present DiCoR, a fully distributed protocol for CPR. Our protocol does not assume any a priori knowledge of the network topology or peer losses, and is resilient to dynamic network changes due to node mobility or the continuous joining and leaving of peers. Detailed simulation experiments, under realistic loss models and network conditions, demonstrate that DiCoR presents a viable solution for timely out-of-band loss repair of MBMS real-time broadcast. I. I NTRODUCTION The point-to-multipoint mode of Multimedia Broad- cast/Multicast Service (MBMS) [1] allows simultaneous distri- bution of identical content to multiple users in 3GPP UMTS cellular networks. Expected applications of MBMS include traffic telematics, news broadcast, music/video streaming, sports replay, and file sharing [2]. As opposed to dedicated point-to-point connections, the MBMS point-to-multipoint mode dictates that the coding scheme and transmit power are chosen in advance for a target set of users, implying a lower level of QoS for users with worse receiving conditions [3]. Previously proposed solutions to improve MBMS reliability mainly focus on application-layer Forward Error Correction (FEC) [4]. However, FEC falls short of fixing all errors for heterogeneous users. Using more complex FEC incurs expen- sive overhead in terms of 3G bandwidth. A possible solution is to have the MBMS base station rebroadcast lost packets. Such retransmission based solutions are challenged by the well known feedback implosion problem [5]. Blind retransmissions, i.e., rebroadcasting all packets, or employing strategies [2] that evade feedback implosion have been proposed. Their utility is undermined by consumption of additional 3G bandwidth [6], which is scarce and expensive. Moreover, nodes having lost packets due to poor local receiving conditions also stand to loose the retransmitted packets. This paper considers Cooperative Peer-to-Peer Repair (CPR) for out-of-band repair of 3G broadcasting losses. CPR is motivated by the widespread availability of multi-homed Cellular connections (for broadcasting) MBMS Service Area IEEE802.11 Ad-hoc Connections ( for CPR repair) Streaming Clients with double WIC Base Station (MBMS-enabled) Fig. 1: Cooperative Peer-to-Peer Repair mobile devices having both 3G cellular and IEEE 802.11 wireless interfaces [7, 8]. Such users can leverage IEEE 802.11 peer-to-peer connections to achieve out-of-band repair of 3G broadcasting losses, as depicted in Fig. 1. This paper presents a fully distributed protocol for CPR. The challenge for such a protocol is to schedule repair transmissions in the presence of interference between wireless peers [9]. Furthermore, dis- tributed nodes are not readily aware of their surrounding net- work topology and distribution of peer losses. Dissemination of such control information, required to guide the scheduling of repair transmissions, incurs overhead. The challenge for a distributed protocol is to realize an efficient tradeoff between this overhead and repair effectiveness. In our short paper [10], we presented an exhaustive search algorithm that assumes availability of global state information to compute the optimal schedule for repair transmissions. [10] uses the exponential-time algorithm for tractable problem sizes to suggest heuristics and a design framework for a distributed algorithm. This paper presents a comprehensive specification of a completely distributed protocol that is derived from that design framework. [10] also presents a very rudimentary performance analysis using a simple protocol specification, and using simplistic broadcast loss models while ignoring the interaction of the 802.11 MAC layer with the protocol. We evaluate and configure DiCoR using detailed packet-level simulations under realistic conditions and interactions with the 802.11 MAC layer. Our major contributions are twofold: • We propose a distributed protocol Distributed Cooperative Repair (DiCoR). DiCoR incorporates the heuristics identified in our short paper [10]. DiCoR dynamically discovers network