Papers

Computer networks and Logistics systems are two very rich fields of study that have grown almost entirely separately since they deal with entirely different entities – information packets vs. real commodities. Recently Physical Internet has been studied that attempts to revolutionize product distribution logistics by emulating the cyber Internet. While the architecture of physical Internet introduces a number of key characteristics (standardization of labeling and packaging, sharing of physical distribution infrastructure among multiple companies, worker friendly logistics etc.), the requirements of perishable logistics and their fresh delivery remained untouched. The distribution of perishable commodities such as fresh food, perishable pharmaceuticals, blood, etc. brings in some special challenges and opportunities that make comparison with information networking particularly apt.

In this paper, we show that considerable synergies exist between Information Networks (IN) carrying time-sensitive information and Perishable Commodity Distribution Networks (PCDN), which can be exploited for solving complex problems in both fields. PCDN involves a flow of packages or package containers (that we can regard as packets) from source (e.g., farm, factory, blood bank, etc.) to destination (e.g., retailer, hospital, etc.). The flows may pass through some intermediate distribution centers that “store and forward” the packets much like IN routers. Most flows also have Quality of Service (QoS) constraints in terms of delivery times and/or flow bandwidth (volume delivered/day). Other than that perishable goods often deteriorate in quality (or freshness) and value as a function of flow time (and other parameters such as temperature, vibrations, etc.) although some perishable goods, such as blood, have a fixed expiry duration. Similarly IN packets often have fixed deadlines, but there are scenarios where the value of information decays steadily with the delay incurred, such as in sensor networks or in financial transactions etc.

However, there are important differences as well. The most fundamental characteristic of a physical packet (or package) is that it is “unclonable”; it can exist only in one place at a time – even though one could surely replace a lost packet by an identical one from the source. Another fundamental difference is that unlike IN, physical packets do not move by themselves; instead, they need one or more additional resources for successful transit. The most important resource is a carrier, which could be a truck, railcar, plane, boat, etc. and the associated driver (unless the carrier is self-driven). Other resources include containers (perhaps even containers within containers), and load-unloading equipment. Although IN systems sometime consider circulation of empty frames that are filled up as the frame passes a sender node, this is rare. INs may also need other resources (buffers, transmission and processing capabilities etc.), but the functionalities are often far simpler. In particular, in PCDN the resources such as trucks, containers, drivers could be distributed throughout the network and need to be properly positioned, whereas the IN resources are generally non-mobile.

In spite of some fundamental differences between IN and PCDN described above, we believe that there is considerable value in attempting to capture the essence of both in a single model. Towards this end, we have explored a 5 layer networking model that encompasses both IN and PCDN and allows application of ideas and techniques across two very different fields. The first layer is the “Physical Layer” that deals with the actual movement of a packet along a media segment or channel. The second layer is the “Media Switching Layer” that provides the media/channel selection, media bridging, and switching functionalities. Then comes the “Routing & Distribution Layer” which supports end-to-end transfer of packets by handling packets at and across distribution/routing nodes. The fourth layer is the “Transport/Delivery Layer” that concerns the end-to-end assured delivery of individual packets (which may have been bundled recursively before transportation and then unbundled for final delivery). The destination will check the packets for loss, damage, deadline expiry, and quality degradation, and accordingly make decisions regarding reorder or replacement. Finally the job of the “Virtualization Layer” is to share the network capacity efficiently while still ensuring isolation among the various services/applications. In particular, this layer can define and maintain one or more virtual networks that are then mapped on to the physical network.

The layering allows us to introduce modeling simplifications via level-specific abstractions. As the automation in PCDN increases, the layered architecture becomes more and more important as it regularizes the product handling at various points. In situations where layering hinders efficient operations, cross-layer methods can be exploited to address them while still limiting the overall complexity. In this paper, we illustrate how such a view can be useful in exploiting the synergies, and expect that it will lead to much broader collaboration, cross-pollination, and unique insights that will significantly advance both fields.

Given the complexity of packet transit in such a unified model, its mathematical modeling is quite challenging and goes well beyond the simple queuing theoretic modeling that is quite common in IN. In particular, such modeling not only needs to deal with batch transmission (or bundling/unbundling), but also with allocation/deallocation of multiple resources whose scope often extends to the entire network instead of being limited to a node or link. The contention for resources results in the “blocking” phenomena, which can be quite difficult to model. For example, a transit may be blocked waiting for arrival of additional packets (to satisfy batching requirements), a suitable number of containers, and a carrier. The containers and carriers may in turn be held up elsewhere in the network. Thus approximate solution methods are almost mandatory, and developing an approximation technique and characterizing its properties becomes quite challenging. In this paper we propose an analytical modeling of such an approximate scenario using the idea of batch queuing and analyzed the impact of waiting time latency of the packages for resources (trucks) on the freshness delivery quality of these packages. We also validate the correctness of our analytical modeling with extensive simulations. We expect that the paper will motivate researchers in the two communities to exploit further synergies and thereby advance both fields.