The quantification of H&S system results in a truckload trucking environment is first achieved using the HUBNET simulator, which is described in Taha (1993), and in Taha and Taylor (1994). HUBNET is a knowledge-based, user-friendly simulation system written primarily in the SIMNET II simulation language with a C-shell user interface. More information regarding SIMNET II can be found in Taha (1988).

The experimental design in Taylor et al. (1994) emphasized the determination of appropriate hub locations, the determination of the number of hubs needed to support North American operations, and decisions regarding permissible driver usage heuristics. Findings indicate that limited H&S implementation may be the best H&S alternative in truckload trucking. For this reason, the experimental focus in this paper is on issues that support limited implementation. In the next section, this experimental design is presented. Subsequently, the results of experimentation within the experimental framework are presented and discussed. The implications of this research in terms of H&S design alternatives in truckload trucking are also presented, along with a general discussions of associated business strategies to support limited H&S implementation.

II. Experimental Design Issues

In this section, we discuss the experimental design issues that affect this research. We begin this discussion of experimental design considerations with a brief summary of experimentation previously completed using the HUBNET simulator. We discuss the development of a baseline scenario for additional testing, and present an experimental design for this additional testing and sensitivity analysis. Also in this section, we present an experimental coding scheme and discuss performance measures.

Experimental Design for Previous Research

As stated above, the primary experimentation completed to date focused on the determination of hub locations, the determination of the number of hubs needed to support H&S operations, and the development of decisions regarding permissible driver usage heuristics. This three factor experimental design was effective in terms of helping to identify critical H&S characteristics and in terms of identifying operational concerns and areas for future research (Taylor et al., 1994).

Hub location methodologies included distance-based hubbing, flow-based hubbing, and hybrid hubbing. Distance-based hubbing attempts to place hubs in locations that are convenient to one another in terms of driving distance. Flow-based hubbing attempts to place hubs in locations such that the absolute value of the number of loads into a service area minus the number of loads out of a service area is minimized. The hybrid methodology involves some consideration of each of the two above factors along with additional considerations as outlined in Taylor et al. (1994).

The number of hubs examined in previous experimental designs were fixed at two levels; 24 hubs and 32 hubs. The 24-hub scenario represents the minimal coverage required to allow adequate H&S coverage in the North American study region. The 32-hub scenario is representative of a larger network that permits additional coverage in key areas.

The driver usage rules examined to date include allowing H&S lane drivers to travel from the home hub according to a star topology with subsequent loads forcing that driver back to his or her home hub. Other rules, which permit longer tour lengths, were also examined.

The Establishment of an Initial Baseline for Comparison

In Taylor et al. (1994), H&S scenarios were compared with one another, and were also compared with a performance baseline established by the point-to-point methodologies prevalent in the truckload trucking industry. For convenience, a coding scheme was developed to aid in identifying scenarios within the factorial design. The notation (HLM/#H/TL) represents a scenario with hub location methodology (HLM) which is equal to D for distance-based, F for flow-based, and H for hybrid; number of hubs (#H) equal to 24 or 32, and permissible driver tour length (TL) equal to 1 or 2 hubs from home. For example (F/24/2) represents a scenario in which 24 hubs are placed according to the flow-based method. Drivers are permitted to travel two hubs from home.

Following extensive experimentation, the (H/32/1) scenario was selected as the initial baseline scenario for additional testing based upon performance relative to five major performance criteria; lane driver tour length, local driver tour length, average miles per driver per day, first dispatch empty miles as a function of trip miles, and average circuity as a function of trip miles. Table 1 presents a comparison of the twelve initial H&S scenarios in terms of five performance ratios obtained by dividing the scenario results for each measure of performance by the best observed results. The (H/32/1) scenario achieves the best results among the candidates in terms of lane driver tour length and in terms of first dispatch empty miles. It is only one percent weaker than (F/32/1) and (F/32/2) in terms of circuity. On the negative side, local driver tour length is 28% higher than (H/32/2), yet still acceptable, and it achieves only 77% of the miles per driver per day as does (F/24/1). This is not especially disturbing given that many more miles in the (F/24/1) scenario are deadhead or circuitous miles.

********Insert Table 1 Here********

This baseline is established as a basis for additional experimentation because of the considerable computational requirements of the HUBNET system which prevent full factorial experimentation with large numbers of factors and levels. In this paper, we now present the experimental design for extended testing and sensitivity analysis, which uses the (H/32/1) scenario as a baseline.

Experimental Design Issues for Extended H&S Testing

All experimentation described in this paper makes use of 32 hubs, allocated according to the hybrid hubbing methodology. Furthermore, all experimentation described in this paper makes use of the one-hub driver tour length rule as specified in the (H/32/1) scenario. Because Taylor et al. (1994) indicate that limited H&S implementation may be the most viable alternative in truckload trucking, the experimental design for extended testing focuses on issues related to limited implementation; allowable circuity, and driver availability. Limited H&S implementation could actually mean several things. It could mean that a full network exists, with only some percentage of total loads flowing through the hub system. It could also mean that a partial (disjointed) network exists, with a large percentage of loads between two or more hubs in a sub-network travelling along heavily used freight lanes. Throughout our discussion of results, it is the former type of network that is being examined. Subsequent to our discussion of results, we will discuss the latter in terms of possible implementation strategies.

Allowable circuity is perhaps the most important factor in terms of limited H&S implementation because allowable circuity is the factor that determines the percentage of loads to be carried via the H&S network as opposed to being carried point-to-point by non-network drivers. The actual point-to-point distance between the pick-up point and the drop-off point is called "trip miles". Circuity, in this paper, is defined as miles in excess of trip miles, and is specified in terms of the percent of trip miles. For example, 10% allowable circuity indicates that a network load is allowed to travel 1.1 times the point-to-point trip miles. If a given load exceeds the allowable circuity as a percent of trip miles, that load is not handled via the network. Instead, a non-network driver is dispatched to move the load according to a traditional point-to-point trucking methodology. Figure 1 shows an example of allowable circuity. In Figure 1, two loads are to be picked up in Denison, TX. One load is to be delivered in Brinkley, AR, and the other is to be delivered in Jackson, TN. The trip miles for these tours are indicated on the figure. Suppose that the Denison, TX location is served by a hub in Dallas, TX. Both destination locations are served by a hub in Memphis, TN. Under the (H/32/1) baseline scenario which allows 60% allowable circuity, the Jackson, TN load would be moved via the H&S network while the Brinkley, AR load would be a point-to-point load. The allowable circuity obviously can play a key role in determining the percentage of network loads.

********Insert Figure 1 Here********

The purpose of Table 2 is to show the effect of allowable circuity upon the percentage of network loads. Consider, for example, the 60% level of allowable circuity from the (H/32/1) baseline configuration. This level of allowable circuity permits 74% of the loads to travel via the H&S network (3711 of 5000 loads, on the average). Note that this level of network participation is not achieved for 24-hub scenarios until allowable circuity reaches approximately 85%, because service areas are much larger in 24-hub scenarios. All loads that are originated and delivered in the same hub service area are automatically designated as non-network loads by the HUBNET system. This is indicated by the "maximum" allowable circuity entry found at the bottom of Table 2.

********Insert Table 2 Here********

The information in Table 2 helps us to establish experimental levels for the circuity factor within our experimental design. Because the (H/32/1) baseline scenario fixes allowable circuity at 60%, it is desirable to perform sensitivity analysis around this figure. For this reason, additional experimentation is performed at the 50%, 60%, and 70% levels. Because the experimentation in Taylor et al. (1994) suggests that limited H&S implementation seems to be the best alternative in truckload trucking, examination of allowable circuity above 70% is not especially interesting. Also, as indicated in Table 2, gains in terms of the percent of network loads come slowly above 70% circuity, which permits 79% of loads to be moved via the H&S network. The examination of circuity levels below 50%, however, is very interesting. Therefore, additional HUBNET runs are completed at the 30% circuity level. Experimentation below this level results in performance in terms of lane driver tour length that is unacceptable relative to the objectives of the study. In summarizing the allowable circuity experimental factor, the levels of allowable circuity are experimentally fixed at 30%, 50%, 60%, and 70%, resulting in 53%, 69%, 74%, and 79% network loads, respectively.

The second major experimental factor examined in this paper is that of driver availability. The reason for examining this experimental factor is to determine whether or not H&S networks would be better served by having more or less drivers available to support operations. This experimental factor is set at three levels, 0.9, 1.0, and 1.1, where the ratio of 1.0 indicates the same number of drivers used in the point-to-point methodology and in the (H/32/1) baseline scenario. The 0.9 level is performed to determine if the miles per driver per day criterion can be improved in comparison with (H/32/1) without adversely affecting other performance criteria. The 1.1 level is examined to determine if the availability of more drivers would better serve the network and lead to further reductions in tour length. The 1.1 level is particularly interesting in recognition of the fact that truly limited H&S implementation requires the maintenance of two very different scheduling and dispatching systems. Pilot runs using driver availability levels above 1.1 lead to poor performance in terms of miles per driver per day. The driver allocation issue is not an experimental factor in this paper. Allocation is based upon the freight density as established in Taylor et al. (1994).

A coding scheme has been developed to aid in identifying scenarios of experimental interest in this paper. Let (AC/DA) represent a scenario with allowable circuity (AC) equal to 30%, 50%, 60%, or 70%, and with driver availability (DA) of 0.9, 1.0, or 1.1. For example, (30/0.9) represents a scenario which allows up to 30% circuity for network loads, but allows only 0.9 times the drivers available in (H/32/1). Because all scenarios in this paper are based on (H/32/1), this indicator is not used in subsequent scenario descriptors.

Measures of Performance

The major performance measures used to establish the baseline performance model were, lane driver tour length, local driver tour length, average miles per driver per day, first dispatch empty miles as a function of trip miles, and average circuity as a function of trip miles. In this paper, the same five measures are considered to be important, yet first dispatch empty miles play a much smaller role in this paper than in Taylor et al. (1994). The reason for this is because HUBNET calculates first dispatch empty miles in a very generic way in a pre-simulation module according to an assumed first dispatch location distribution from each hub. In other words, first dispatch empty miles are assumed to originate at the hub instead of the last drop-off location for local drivers. This is a useful means of rapid approximation, and is a valid means of comparing scenarios with identical network load distributions such as those tested in Taylor et al. (1994). However, this is not a valid means of comparing scenarios with different network load distributions. In the experimental analysis presented in this paper, changing AC also changes the network load distribution and prevents a valid direct comparison. For this reason, first dispatch empty miles is presented as a secondary criterion in this study, with an examination of performance levels in two extreme cases; the current point-to-point system, and the 100% AC case.

Verification and Validation

As in Taylor et al. (1994), the validity of supporting load data is guaranteed by using actual data supplied by J.B. Hunt, Inc. Because HUBNET has been in use for more than one year at J.B. Hunt, Inc., and at the University of Arkansas, the code has been fully verified. Moreover, the results obtained using HUBNET seem consistent in terms of observed trends and seem reasonable relative to expert judgement. A replication design is used to ensure the independence of data between runs for statistical testing purposes.

III. Analysis of Results

In this section, we discuss the results of the experimentation described in the previous section. This includes a discussion of numerical results obtained for the four primary performance criteria and one secondary performance criteria discussed previously. This section also includes statistical information in the form of Analysis of Variance (ANOVA) testing. HUBNET computational performance is commensurate with that described in Taylor et al. (1994).

Performance Evaluation and ANOVA Testing

Twelve major scenarios are evaluated in this paper. Recall that four levels of allowable circuity and three levels of driver availability are considered. In this section, the results of this performance evaluation are presented. Each scenario consists of five independent replications of 5000 loads each. As in Taylor et al. (1994), the primary performance measures of miles per driver per day and average circuity miles are considered proprietary by J.B. Hunt, Inc. Therefore, performance relative to these criteria are presented as a percentage of current point-to-point methods and as a percentage of trip miles, respectively. The secondary performance measure of first dispatch empty miles is also measured as a percentage of total trip miles.

The results presented in Figure 2 are largely intuitive. Simply stated, the results seem to indicate that as allowable circuity increases, lane driver tour length decreases. The reason for this phenomenon is that as allowable circuity increases, the percentage of loads moved via the H&S network increases. Greater load volume and more widely distributed loads provide much greater opportunity in terms of dispatching possibilities so that drivers can return to their home hub with greater frequency. This intuitive result is supported by the ANOVA results presented in Table 3. The ANOVA testing indicates that the AC factor significantly affects lane driver tour length at an alpha level of 0.0201. Neither the DA factor or the AC/DA interaction effect are statistically significant at an alpha level of 0.05.

********Insert Figure 2 and Table 3 Here********

Sharp increases in lane driver TL occur when AC is reduced below 0.5 to 0.3, because more loads are forced to travel as point-to-point non-network loads. Even so, TL is much smaller than in point-to-point methods as reported in Taylor et al. (1994). The reader should recall that even with AC at 30%, more than 50% of all loads travel on the H&S network. The fact that lane driver TL is still quite small compared to point-to-point methods is therefore not disturbing experimentally, and even provides some level of validation for HUBNET results. HUBNET, after all, places a great deal of emphasis on TL reduction for network loads. The fact that the TL increases are much steeper for low circuity is also encouraging.

The general performance is also intuitive in terms of the driver availability factor. Without exception, those scenarios with a smaller number of drivers available have longer tour lengths than the corresponding scenarios with a larger number of drivers available. As we will demonstrate subsequently in our discussion of local driver TL, this is more an effect caused by local pick-ups and deliveries than one caused by lane driver deliveries. Less intuitive is the unexpected increase in lane driver TL between the (60/0.9) and (70/0.9) scenarios. This phenomenon will also be described in our discussion of local driver TL.

Figure 3 demonstrates the effects of circuity and driver availability on local driver TL, given the current means of allocating drivers among lane, local, and non-network alternatives based on load density. Note that the TL performance is opposite from that of lane drivers. Specifically, as circuity is decreased, local TL also decreases. Local driver TL is greater for those scenarios in which driver availability is lower, because backlogged pick-up requirements increase the probability of local drivers getting two-way loads to and from the hub.

********Insert Figure 3 Here********

The shape of local service areas for H&S pick-ups and deliveries is user defined through dialogue driven by HUBNET's C-shell user interface system. The effective size and shape of the service area, however, is heavily dependent upon the AC value selected. Consider, for example, the scenario depicted in Figure 4. For small AC values, loads will generally not be considered for H&S network travel unless they are either very close to the hub or unless they are in more or less a direct line behind the hub with an existing spoke or lane. In these cases, local pick-ups would be permissible because the load would potentially need to pass near the hub location even using a point-to-point system, thus reducing total circuity. Consider, for example, load location "A" and load location "B" in Figure 4. Even though both are equidistant from the hub, load A is much more likely to be a network candidate than load B for small AC values. This is primarily because loads going in the general direction of Lane 3 could be delivered with minimal circuity. Network loads originating at load location B would need to travel to the hub prior to beginning "productive" movement if travelling along lane 1 or lane 2 and twice the distance to the hub if travelling in the general direction of lane 3. As AC increases, the size and shape of the effective service area changes, and the distribution of network loads changes.

********Insert Figure 4 Here********

The net effect of increasing AC is twofold from the viewpoint of local driver TL. First, the number of loads farther from the hub increase, drawing local drivers farther and farther from the hub for deliveries and pick-ups. Secondly, this increased driving distance and load availability increases the probability that drivers will have a return load to the hub. This is encouraging in terms of deadhead elimination, yet does add slightly to local driver TL. Even so, this trade-off seems to be a good business practice. This trade-off seems even more valid when the results of ANOVA testing are considered. In Table 4, these results are tabulated. The results indicate that while interesting in terms of observing trends in performance data, these trends are not statistically significant. In other words, we can apparently increase loaded miles without adversely affecting local driver TL significantly.

********Insert Table 4 Here********

If we were to examine these results in isolation, we might also reach the conclusion that driver availability is not an important issue. It is when we examine lane driver tour length concurrently with local driver tour length that we gain insight into the importance of driver availability. Note in Figure 2 that when DA is equal to 0.9, a sharp increase occurs as we shift from 60% to 70% AC. This increase is brought about because at high AC values, local driver tour length increases such that they are no longer able to feed lane drivers at the hub adequately. Pilot studies with larger AC values show similar trends at DA values of 1.0 and 1.1. Eventually, even for higher DA rates, the lane driver TL benefits associated with higher AC are overcome by the problems associated with local driver feeding rates. The threshold for this turning point, however, occurs at progressively higher AC values when more drivers are available.

It is possible that changing HUBNET's allocation procedures could have kept the performance curves presented in Figure 3 flatter in terms of local driver TL as AC is increased. Specifically, as AC is increased, it would appear that HUBNET's driver allocation procedure could possibly be modified to allow additional local drivers and fewer lane drivers. This examination, however, exceeds the current experimental scope and is suggested as a primary component of future research strategies using the HUBNET simulator.

Table 5 presents the results of ANOVA testing for the miles per driver per day criterion. Interestingly, neither AC, DA, or their interaction with one another significantly impact the observed values of miles per driver per day. Figure 5 presents these results graphically compared to the point-to-point baseline. These figures are presented in percentages to disguise the proprietary baseline information provided by J.B. Hunt Transport, Inc.

********Insert Figure 5 and Table 5 Here********

The percentage of miles per driver per day that are considered productive are affected by the circuity and first dispatch empty miles criteria. Once again, these criteria are considered proprietary by J.B. Hunt Transport, Inc. Comparisons with these criteria are therefore made on the basis of circuity and first dispatch empty miles as a percentage of total trip miles. Obviously, allowable circuity will heavily impact the circuitous miles. Figure 6 demonstrates this result. Because circuity is a static parameter calculated in the HUBNET pre-simulation function based on hub and service area locations relative to loads, driver availability can have no effect. For this reason, ANOVA testing for this performance criterion is not especially interesting. However, we have performed a number of different multiple means comparison tests to determine if statistical differences exist in the mean values obtained for circuity for various AC values. Some of the tests performed, such as Duncan's Multiple Range Test, control the comparisonwise error rate. Others, such as Tukey's Studentized Range Test, control the experimentwise error rate. In all cases, each test indicates that there is a highly significant difference between each of the mean circuity values for all levels of AC. Using point-to-point methods, average circuity is 3.5% of trip miles. This is brought about by the necessity to return drivers to their homes. Figure 6 demonstrates that HUBNET data is trending correctly toward the point-to-point baseline value as AC is reduced.

********Insert Figure 6 Here********

As stated earlier, the first dispatch empty miles criterion has been dubbed a secondary performance criterion in this study because HUBNET does not permit valid direct comparisons for this measure of performance between scenarios which feature the different network load distributions brought about by circuity changes. For the point-to-point system, first dispatch empty miles are approximately 5.6% of trip miles. At 100% allowable circuity, first dispatch empty miles increase to 8.1% of trip miles. Recall from Table 2 that 100% AC results in approximately 87% of all loads being carried by the network. The reason for the first dispatch empty miles performance degradation for higher AC values is that more network loads means more network drivers. Network drivers terminate their tours at central hub locations. The point-to-point system results in a wider dispersion of available drivers within the service area. Moreover, because areas of high delivery volume are also normally associated with areas of high load origination volume, driver ending locations are likely to be near freight pick-up locations. The result is that first dispatch empty miles are likely reduced when an increasing proportion of drivers are travelling outside of the H&S network delivery system.

Comparative Performance

In this section of the paper, we compare the performance of the various scenarios of experimental interest relative to one another. Table 6 is helpful in this regard. For each of the twelve primary scenarios, five ratios are used to measure performance. The ideal scenario would have a ratio of 1.00 for all five performance measures. Departures from 1.00 represent increases or decreases from the best observed scenarios. For example, the (70/1.1) scenario performs best in terms of lane driver TL with (30/0.9) having TL values 26% higher. The best scenario in terms of local driver TL is (30/1.1). The best scenario for miles per driver per day is (30/1.0). The 30% AC scenarios provide the lowest values for both circuity and first dispatch empty miles.

********Insert Table 6 Here********

Based on the information presented in Table 6, it would appear that the best observed scenario from this extended testing and sensitivity analysis is the (30/1.0) scenario based on the (H/32/1) baseline. This scenario is the best among H&S scenarios examined in terms of miles per driver per day, average circuity, and first dispatch empty miles. It is only 1% behind the (30/1.1) scenario in terms of local driver TL, yet it achieves this level with 10% fewer drivers. Although it provides for lane driver tour lengths that are 18% higher than (70/1.1), these tour lengths are certainly adequate at less than 50 hours.

Compared to point-to-point delivery methods, the best H&S scenario is very competitive. The (30/1.0) scenario achieves 89.4% of the miles per driver per day achieved using point-to-point methods. Circuity and first dispatch empty miles are 6.74% and 6.35%, respectively, compared to 3.5% and 5.6%, respectively, under point-to-point delivery rules. Furthermore, the performance level for the (30/1.0) scenario compared to the (H/32/1) baseline represents improvements in terms of miles per day, circuity, and first dispatch empty miles of 4.7%, 48.4%, and 10.7%, respectively.

The greatest benefits of H&S network delivery systems compared to the point-to-point method is in terms of lane driver TL. Although the (30/1.0) scenario is 26% higher than (70/1.1), it still provides an 88.5% improvement in TL compared to point-to-point delivery systems. Local driver TL values are 96.0% better than average point-to-point TL values.

As in Taylor et al. (1994), the results of this performance comparison seem to indicate that the best possible implementation strategies for H&S networks are limited in scope. The best scenario identified from the sensitivity analysis and extended testing presented in this paper is one in which only 53% of total loads are moved via the H&S network. Relative to some performance measures, even fewer loads should travel via the H&S network. Movement in this direction, however, leads to increasingly rapid performance degradation relative to TL, the primary motivator for hubbing in truckload trucking.

IV. Implications for Operations

The initial findings of Taylor et al. (1994) using the HUBNET system, and the findings of extended testing and sensitivity analysis presented in this paper suggest that limited implementation is the best H&S alternative in the truckload trucking environment. It is not clear at this point, however, how truckload trucking companies could best exploit the opportunities presented in these papers. As discussed earlier, the best means of limited implementation is not entirely clear. Truckload companies could implement a full H&S network with limited traffic or could implement a partial network with heavier traffic. In this section, we will briefly discuss both alternatives along with implementation plans currently being considered within the truckload trucking industry.

Complete Networks With Limited Traffic

We shall begin this discussion with the implementation of complete networks with partial load participation, because this has been the primary focus of work completed at this time using the HUBNET system. So far, the best H&S system identified makes use of 32 North American hubs with restrictive driver rules in terms of tour length. Placing hubs according to flow-based and distance-based rules seems to lead to the best performance. Networks that carry approximately 50% of all loads seem to perform well with roughly the same number of drivers as required in point-to-point delivery systems. This does not imply, however, that better network configurations or control rules do not exist. An obvious extension to this work would be to focus on driver allocation rules in an effort to better feed lane drivers. Load assignment logic could be enhanced to study the impact of allowing lane drivers to make selected local pickups and deliveries if they are within certain circuity limits. Also, any hub location methodology used by industry must consider the demographics of the prospective region to ensure that an adequate supply of truck drivers exists to staff the proposed hub.

J.B. Hunt Transport, Inc. is experimenting with a zone delivery plan that could become an important first step in H&S implementation. Although this delivery system does not actually make use of hubbing, it does restrict drivers to a zone similar to, yet larger than HUBNET service areas. Basically, the zone system ensures that drivers are able to return home for at least one day each week. Real-time optimization technology is used to identify load switches between drivers that will keep each driver within their zone and returning home at the right time. This decision support system identifies beneficial load switches, along with recommended switch points and times, based on current load positions and final destinations. In a sense, this allows a H&S network to be implemented with a nearly infinite number of hubs since truck stops, rest areas, and existing terminal yards are used as switch points, or hubs. The large number of virtual hubs used helps to minimize the circuity and first dispatch empty miles incurred in the zone delivery system.

The zone delivery experimentation is too new to be able to determine the long term effects on driver retention, yet initial feedback is encouraging. Many of the zone design issues are the same as those that affect H&S system design. Specifically, it is known that not all drivers can be zone drivers, yet the optimal percentage of drivers to designate as zone drivers is currently unknown. Also, how can zones (service areas) be configured to eliminate deadhead miles? Where should zone drivers be permanently stationed to eliminate deadhead? What freight mix should be solicited to complement the H&S network implied by the zones? H&S solutions would undoubtedly encounter the same types of growing pains as optimal participation levels and system configurations are determined.

Partial Network Configurations

The second H&S implementation alternative has not received as much attention, yet may prove to be the most viable option available to the truckload trucking industry. This option is that of creating disjoint partial networks for consideration. In this type of configuration, it may be possible to identify two high load density areas in cities that are conveniently located relative to one another in terms of driving distance. Consider for example, the Dallas, TX and Memphis, TN hubs depicted in Figure 1. The hubs are located a convenient one-day drive from one another at 456 miles. Both of these hubs represent high volume load density areas around major United States cities. It may be possible to make regular runs down their connecting lane. These runs could be for traffic originating or being delivered in the Dallas or Memphis areas, or could move loads that are passing through this lane on the way somewhere else. In either case, the lane structure implies that appropriate supporting hub service areas are in place. This complicates issues profoundly, but less so than if a full network were to be established concurrently.

Implementing a partial network avoids the risky and potentially fatal conversion to a new business process. Lane structures could be implemented incrementally, either by adding service to new markets with the necessary freight characteristics, or by converting selected, existing markets like Dallas/Memphis to a H&S structure. Either method allows for an evolutionary change process, and avoids the risks associated with an immediate implementation of a full H&S network.

It may be possible to even make some percentage of truckload trucking business more regular and scheduled, as in airline travel. For example, if the number of Dallas-Memphis lane drivers were carefully managed, it would be possible to sell regular route capacity in much the same way that point-to-point capacity is sold. This could become a new marketing niche for companies that exploit this idea effectively. The idea offers a win-win solution to the problems of driver shortages faced by trucking companies and the freight handling capacity shortages faced by industry. Drivers would be attracted to the regular work schedule, while shippers could reserve scarce capacity well in advance with the carrier of their choice.

De facto partial networks are being formed within the truckload trucking industry via increasing intermodal integration with rail networks. Essentially, fixed rail lines are lanes, and rail terminals are hubs. Local service areas feed these hubs. As intermodal traffic increases, the truckload trucking industry is moving toward limited H&S configurations by default. J.B.

Hunt Transport, Inc. is currently handling the growing pains associated with intermodal business, and is evolving business policies to effectively manage this integration.

V. Concluding Remarks

In this paper, the authors present the results of sensitivity analysis and extended testing associated with experimentation that originally appears in Taylor et al. (1994). The findings of this testing indicate that limited H&S implementation appears to be the best H&S alternative in the truckload trucking industry. The primary motivator for hubbing in this unlikely industry remains that of driver retention via shorter tour lengths. Viable H&S configurations to reduce TL are presented. These configurations result in much shorter driver tour lengths than traditional methods of dispatching, yet do not perform as well relative to other important criteria such as average miles per driver per day.

Future work is needed to support limited H&S implementation. Slowly, we are starting to understand how these networks should be designed. Even so, we have not answered all configuration questions. For example, we still have not evaluated driver allocation rules to support specific H&S applications. More importantly, we have not fully answered the important question of how we go about implementing H&S strategies without massive short term disruption of delivery systems. These questions should provide researchers with a very interesting set of challenges in the near future.

VI. References

Braklow, J.W., Graham, W.W., Hassler, S.M., Peck, K.E., and Powell, W.B. (1992), "Interactive Optimization Improves Service and Performance for Yellow Freight System", Interfaces, Vol. 22, No. 1, pp. 147‑172.

Kanafani, A., and Ghobrial, A.A. (1985), "Airline Hubbing--Some Implications for Airport Economics," Transportation Research, Vol. 19A, pp. 15‑27.

Mele, J. (1989), "Carriers Cope With Driver Shortage," Fleet Owner, Vol. 84, No. 1 (Jan) pp. 104‑111.

Mele, J. (1989), "Solving Driver Turnover," Fleet Owner, Vol. 84, No. 9 (Sept.) pp. 45‑52.

Morrison, S.A., and Winston, C. (1986), Economic Effects of Airline Deregulation, Brookings Institute: Washington, D.C., USA.

Taha, H.A. (1988), Simulation Modeling and Simnet, Prentice-Hall, Inc.: Englewood Cliffs,

NJ, USA.

Taha, T.T. (1993), " An Integrated Modeling Framework for Evaluating Hub-and-Spoke Networks in Truckload Trucking Operations, " Master's Thesis, Department of Industrial Engineering, University of Arkansas, USA.

Taha, T.T., and Taylor, G.D., (1994), "An Integrated Modeling Framework for Evaluating Hub-and-Spoke Networks in Truckload Trucking Operations," The Logistics and Transportation Review, Vol. 30, No. 2, pp. 141‑166.

Taylor, G.D., Harit, S., and English, J.R. (1994), "Hub and Spoke Networks in Truckload Trucking: Configuration and Operational Concerns," In review for publication in The Logistics and Transportation Review.

Table 1. Performance for Major H&S Scenarios (Taylor et al., 1994). _____________________________________________________________________________

Measure of Performance*

Scenario (1) (2) (3) (4) (5)

_____________________________________________________________________________

(H/24/1) 1.18 1.48 0.89 1.17 1.35

(F/24/1) 1.06 2.49 1.00 1.42 1.40

(D/24/1) 1.15 2.33 0.97 1.47 1.84

(H/24/2) 6.20 1.29 0.83 1.17 1.35

(F/24/2) 5.65 1.98 0.92 1.42 1.40

(D/24/2) 6.63 1.85 0.97 1.47 1.84</