Minutes of the Washington D.C. Manet WG Meeting Reported by M. Scott Corson 1. Introduction The manet WG met in two sessions. The meeting began with the standard agenda bashing session, which was then followed by presentations and discussions covering various topics including: updates to existing drafts for DSR, AODV and TORA; presentations of new drafts on STAR, RDMAR and INSIGNIA; a presentation of a Nokia Link API; and a presentation of a Nova Engineering radio/router product built with downloadable support for multiple manet protocols. In addition to the central discussions involving individual protocols, some time was spent on issues such as layering and modular design. A Link Application Programmer Interface (API) was presented by Nokia. It presents an approach for separating link and network layer functionality in a reusable fashion. Also, the presentation on INSIGNIA discussed the separation of concerns regarding routing, Quality of Service (QoS) signalling and resource reservation/prioritization at the MAC layer for manet protocols. While QoS support is not the WG's near term focus, the discussion of the layering approach set forth in INSIGNIA was constructive. The following sections give more details regarding various meeting presentations. 2. Dynamic Source Routing Protocol Dave Maltz from Carnegie Mellon University presented recent work on the Dynamic Source Routing (DSR) protocol. The work is focused in two directions: the use of path identifiers to reduce source routing overhead, and an alternative route cache data structure for storing learned topology data. Both approaches rely on DSR's route-request/route-reply exchange to learn source routing information. The path identifier approach essentially implements a form of header compression. After having learned a source route, a source can assign a path identifier to the route which is sent in the data along with its associated source route. Nodes along the route then associate the identifier with the source route. Once the source receives a packet from the destination indicating it has received a data packet with the path identifier, the source can begin sending only identifiers in the data packets, thus reducing source routing overhead. The alternative route caching technique builds a partial topology database from which source nodes construct their source routes. The set of route caches throughout the network essentially form a loosely-consistent link state topology cache. In a fashion similar to traditional link state algorithms, DSR uses a Dijkstra's shortest-path computation to compute a shortest-path route to a desired destination. It has been shown in simulation to improve performance relative to the previous route cache structure. The two methods, path identifiers and the alternative route cache, are independent and may be used in isolation or in combination. 3. Ad hoc On-demand Distance Vector Protocol Charlie Perkins from Nokia presented recent work on the Ad hoc On-demand Distance Vector (AODV) Protocol. Notable modifications include creation of a new ``route error" message. This was added because a single link failure would commonly result in loss of routes to multiple destinations at an upstream neighbor. It was noticed via simulation that protocol performance could be improved by creating an explicit message to handle these events. Also, a ``group hello" message was created in the multicast version of AODV. This functionality was previously implemented with a ``route reply". Ns-based simulation results (to appear at INFOCOM 2000) were presented comparing AODV with DSR. These simulations used the familiar 300m by 1500m rectangular topology from the original CMU Mobicom '98 paper with the same 802.11 MAC layer model. Simulation results were obtained for 50 and 100 node networks using the random waypoint mobility model from the original CMU study. The traffic scenarios were somewhat more challenging than the original CMU study, with 40 CBR sources generating traffic. Throughputs on the order of 70% were reported (in contrast with the 90% and above figures from the CMU study) indicating a greater degree of data traffic and congestion in the network. The general trend of the results reported indicate that DSR is performs better with light traffic loads and less mobility, whereas AODV performed better with heavier traffic loads and higher mobility rates. In the simulations AODV generated more control packets than DSR in small networks, but this number tends to converge with that of DSR as the number of nodes increases. There was significant debate attempting to assess the meaning of these findings in the light of past simulation studies by CMU and Ericsson. 4. Temporally-Ordered Routing Algorithm Vince Park from the U.S. Naval Research Laboratory presented recent modifications to the Temporally-Ordered Routing Algorithm (TORA). Some incomplete sections from the earlier draft were filled in, most notably TORA's reaction to link failures and recoveries. Also, new modes for ``proactive" operation and ``periodic optimizations" were incorporated into the draft. These modes can be turned on selectively on a per-destination basis, under the control of the destination itself. In proactive mode, nodes with non-NULL heights automatically exchange these a new neighbor over a newly-formed link. Periodic optimizations are destination-initiated, and serve to periodically improve the routing data structure. The extent of these optimizations may be network-wide, or they may only impact nodes with non-NULL heights. The effect of these optimizations is to (re)set the heights of the nodes to the all-zero reference level as defined in TORA. 5. Source Tree Adaptive Routing Protocol J.J. Garcia-Luna-Aceves from the University of California, Santa Cruz presented the Source Tree Adaptive Routing (STAR) protocol. STAR is a link state-based algorithm. It disseminates link state information on a ``need to know" basis. To do so, nodes exchange source-based spanning trees. However, unlike previous algorithms, these trees need not specify shortest paths to all destinations. Nodes keep track of their one-hop neighbor sets and the source trees reported by their neighbors, and from these form their own source trees which are exchanged with their neighbors on a need-to-know basis. Two sets of criteria define the need to know---the Optimized Routing Algorithm (ORA) and the Least Overhead Routing Algorithm (LORA). When ORA functioning is turned on, the trees exchanged are shortest-path trees. Under LORA propagation rules, seven criteria determine when nodes update and propagate routing information. The intent of LORA-based propagation is to incur the least possible routing control overhead. In this mode STAR begins to function more like an on-demand routing algorithm than previous table-driven approaches derived from link state technology such as ALP. Simulation results were presented comparing STAR with DSR. The simulations were conducted on the C++ Protocol Toolkit (CPT) Simulator after having ported a ns-based DSR model into the CPT simulator. Promiscuous mode was not enabled for either protocol. Also, routing protocols were not able to re-schedule a packet that had been sent to the 802.11 MAC layer for transmission. Networks of 20 nodes were simulation in a 5000m by 7000m area. Movement rates of 20 m/sec. were simulated using a random waypoint model. Statistics for average nodal degree and frequency of partitions were not given. In comparison with DSR, the study showed that as the number of traffic flows increased, STAR began to outperform DSR. Also, the performance of DSR was seen to deteriorate relative to STAR as movement levels increased. The poorer throughput performance of DSR was attributed to its frequent choice of poor routes, leading to a significant number of ROUTE ERROR packets generated when packets would not be deliverable at the link layer. The distributions of packet delivery latency for the two protocols was similar. 6. Relative Distance Micro-discovery Ad hoc Routing Protocol George Aggelou from the University of Surrey presented a draft on the Relative Distance Micro-discovery Ad Hoc Routing (RDMAR) Protocol. The protocol is structured along the lines of a distance vector algorithm and uses sequence numbers to maintain loop freedom during the route discovery process. A key concept behind RDMAR is its generic approach to the maintenance of active paths (Route Maintenance procedure). The Route Maintenance algorithm in RDMAR is a distributed operation that exploits the spatial relationship of nodes when a failure along an active route occurs and depending on the relative distance of the node that reports the failure from the calling and called nodes, two heuristics are considered: a) if its relative distance from the called node is smaller or equal to this from the calling node, then RDM is to be applied to localise the repair of the failed route on the region of the network where the failure occurs; otherwise, b) b) the node proceeds and informs the calling node about the failure to deliver the call through this path. 7. Nokia Link API At the request of the WG chairs, Chane Fullmer of Nokia presented recent work on a Link Application Programmer Interface (API). The intent of the presentation was to have the WG revisit issues of modular design and layering via the use of extended interfaces. While there has been significant recent interest in developing manet protocols atop 802.11 link layers, it should be remembered that 802.11 is only one of a growing set of link layers over which manet protocols may run. The Nokia Link API separates network and link layer control functionality through the use of a modular, extensible software interface. The API specifies only the external behavior of a MAC layer, not its implementation. The benefits of such an approach are that it permits multiple manet protocols to be implemented on a given MAC layer. Also, multiple MAC protocols---supporting possibly the same interface or a set of interfaces hierarchically derived from a base interface---can be used by a given manet protocol. After the presentation, the WG briefly discussed the pros and cons of utilizing a layered design. The WG chairs feel that the use of standardized interfaces may benefit the the WG's efforts as a whole in the long term. The WG's rough consensus seems to be that while the long term benefits of such an modular approach are desirable, the functional understanding is still too immature to permit definition of a useful interface that would be implemented by wireless vendors and used by wide set of protocol implementors at the present time. However, further maturity and progress in this area is likely overtime and the WG may revisit this issue when considered more mature. 8. INSIGNIA Andrew Campbell from Columbia University presented a draft on INSIGNIA---a protocol for in-band QoS signalling in MANETs. The term ``in-band signaling" refers to the fact that control information is carried along with data in IP packets. This contrasts with explicit out-of-band approaches such as RSVP. The INSIGNIA draft argues that in band signalling is preferable when supporting end-to-end quality of service in dynamic environments such as MANETs. INSIGNIA is designed to support the delivery of adaptive real-time services and includes fast session/flow/microflow reservation, restoration and adaptation algorithms between source/destination pairs. Andrew also discussed how INSIGNIA fits into a broader vision of a wireless flow management model for MANETs and how it interfaces to proposed manet WG routing algorithms. 9. Nova Programmable Manet Radio Mike Geile from Nova Engineering presented the architecture of a readily-programmable manet radio/router product. In its default configuration, the router's IP kernel comes loaded with an early version of the TORA/IMEP protocol suite. However, the kernel's architecture is open, modular, and has been designed to support the operation of multiple manet routing protocols. Protocols coded to the kernel's network API can be downloaded into the radio. This provides manet protocol implementers an alternative to wireless LAN technologies---e.g. the 802.11-based family of radio products---which is optimized for outdoor usage. The radio has a transmission range up to 10 miles in normal operation. 10. Summary of WG State In conclusion, the manet WG continues to move forward in the areas of implementing, refining, and testing routing protocol draft proposals. Varied participants are continuing to report both simulation and ``live" network testing results. Also, additional manet protocol enhancements areas (e.g., link interface, modular design and quality of service) continue to be discussed and presented within the group. Commercial products based on early specifications of manet WG protocols (i.e., at Nokia and Nova) are being sold, which indicates that a market for the technology is beginning to emerge.