We classify related work into: (a) QoS architectures; (b) overlay-based techniques; (c) loss recovery mechanisms.
QoS architectures: OverQoS differs from previously proposed QoS architectures because it does not require QoS mechanisms in all routers in the network. IntServ [10] requires each IP router to implement per-flow admission control on the control path, and per-flow classification, buffer management and scheduling on the data path. Similarly, DiffServ [8,24] requires edge routers to perform per-flow or per-aggregate classification, buffer management and scheduling, and core routers to perform per-class operations.
OverQoS can leverage the service provided by the underlying network to enhance its services. For instance, within a DiffServ domain, OverQoS may use Expedited Forwarding (or premium service [24]) and provide per-flow bandwidth (and perhaps delay) guarantees. In addition, OverQoS can use techniques like the one proposed in the SCORE architecture [35] to improve its scalability, by having only the first OverQoS node on a flow's path maintain state.
To address the scalability problems of providing end-to-end services, several recent papers have advocated the idea of using endpoint measurement-based admission control (EMBAC) [11,20,14]. With EMBAC, an end-host measures the network characteristics of a path and accepts a flow only if the flow's requirements can be satisfied by the path. However, unlike OverQoS, all EMBAC solutions assume that all routers implement some mechanism to isolate the admission-controlled traffic from the best-effort traffic.
Overlay-based Techniques: Several papers have proposed the use of overlay-based approaches for deploying multicast [12,21] and improving routing functionality (e.g., resilience, as in RON [7]). These systems are motivated in large part by the difficulty of modifying the IP layer both in terms of deployment and in terms of system robustness.
Within the context of QoS, edge-to-edge congestion control [18], a proposal to support a limited range of bandwidth services using an overlay framework, also requires modifications at all edge routers in a domain to achieve its functionality. Service Overlay Network [13], is a recent proposal that purchases bandwidth with certain QoS guarantees from network domains using SLAs and stitches them to provide end-to-end QoS guarantees. Such an architecture would still rely on the underlying domains to meet their specified QoS requirements. For streaming audio and video, multimedia proxies offer the services of smoothing losses [34] and selective discard/recovery of packets [37]. While OverQoS can leverage many of these techniques, two issues differentiate these works from OverQoS: (a) OverQoS can apply the same QoS enhancements within the network as opposed to end-to-end; (b) streaming media flows in OverQoS can be shaped as part of a larger aggregate as opposed to being treated as separate flows.
Loss Recovery: FEC and ARQ based approaches have been investigated in the context of packet audio, video and Internet telephony [9]. Since the FEC constraints are different in these applications (recovering a fraction of packets may be sufficient), we may not be able to apply these results directly to our setting. However, classical coding mechanisms used in wireless networks can potentially be applied to our problem [22,31,38].