Andreas Haeberlen: The authors consider the routing problem in delay tolerant networks (DTNs). In a DTN, most links are only available sporadically, and the network may become partitioned frequently. Several simple routing algorithms are proposed and briefly evaluated. The motivation in the paper is extremely far-fetched; the 'remote village' scenario is almost trivial, and I completely fail to see why San Francisco (which is full of top-notch, high-speed networks) would need to form a DTN with its bus system. If DTNs are a problem as rich and challenging as the authors would make us believe, there has to be a better use case! But I suspect the problem here is more fundamental. As the authors have observed, it is difficult to get good performance out of a DTN if only limited information is available. This is why in practice, DTN-style systems (such as the postal system and early BBSes such as Fidonet, which are never mentioned in the paper) are built using a structured network and hierarchical routing. It remains unclear what advantage DTN routing would provide in this case. Several sections of the paper are very vague or could be removed entirely without much loss to the reader (a prime example of this problem is Figure 3). ************************************************************************ Amit Kumar Saha: Title: Routing in Delay Tolerant Network Authors: Jain, Fall, and Patra Sigcomm 2004 ---------------------------------------------------------- Abstract: -------------- The paper formulates the delay-tolerant networking routing problem and suggests certain classes of solutions to the problem. The paper also provides simulation evaluation of these suggested solution classes. Comments: ----------------- Even though the problem is moderately relevant yet the treatise is very childish. The authors motivated the problem with two scenarios: a village like scenario and a city-bus scenario. The proposed solution is not a practical workaround for any of the scenarios. A village does not have network connection because of other socio-economic reasons and those have to be looked at rather than providing routing. Recently such massive efforts have started in Indian villages and many villages have already been provided with full time dialup (56Kbps) connections via wireless long distance connections (something like dish TV network). These Indian companies have also started providing the same solution to countries like Brazil. The scenario of busses running in a city requiring delay tolerant networking is even more unrealistic. All cities in the US and other developed country have cellphone coverage and most cities all over the world are on the verge of having cellphone coverage. Apart from the unrealistic scenariors the paper also has very little contribution. The paper is more suited for a magazine article which gives an idea of different (already existing and mostly obvious) ways of solving a problem. Also, all the approaches present in the paper (apart from the First Contact approach) uses global knowledge which cannot be implemented in a distributed manner in most scenarios. - Amit **************************************************************************** Dan Sandler: Jain et al. define the problem of routing in an environment where certain links are disconnected for long, predictable periods of time. The chief example given is that of network connectivity by non-geosynchronous satellite. Connectivity is characterized in terms of scheduled "contacts" with the link in question. Conventional routing methods fail because they are too sluggish to adapt to brief contacts with links; routes will typically never be found which occupy these links because they will almost never appear to be part of the connected components of the network graph. The authors propose several algorithms to operate in these "delay tolerant" networks; these algorithms are adapted to varying levels of available knowledge about the network, and offer commensurate levels of routing performance as a result. This knowledge is formulated as a set of "oracles" which can answer questions about traffic demand and link availability. To take advantage of sophisticated information about link availability, the authors modify standard graph-searching algorithms to take propagation delay into account when computing the cost for traversal of a given link. The chief difficulties with this paper revolve around (1) the oracles: how can these be built and maintained in a distributed fashion, such as the ad-hoc environments the authors envision? and (2) the problem domain: is it really so common to find non-geosynchronous satellites which may be used for network communication? Presumably as time goes on, the edge of Earth's atmosphere will become crowded with geosynchronous satellites capable of servicing connectivity demands. It is also unclear to me how the algorithms (modified Dijkstra) deals with cost functions which are too complex to be integrable, or which vary unpredictably. It is reasonable to say that if a link's capacity is unpredictable that it falls outside the scope of this paper, but what if a link is merely "late"? Hopefully satellites will never be significantly late (though they may drift due to inaccuracies in the time-variant model) but what about the bike courier described elsewhere in the paper? Data may then be scheduled for links which are not ready. *************************************************************************