Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
the neighborhood. The model suggested two interesting findings. The first was that anyone who makes trips to or from the Tenderloin, for any reason, chose walking, transit, and bicycling at greater rates than the average San Franciscan. Trips to and from the Tenderloin were about half as likely to be made by car as the average San Francisco trip. The other interesting finding was that when non- Tenderloin residentsâ trips were included in the totals, car use increased. This indicated that non- Tenderloin resi- dents who make a trip to the Tenderloinâ for work, social activities, or any reasonâ were one- third more likely to use a car than a Tenderloin resident. This finding suggested that about one- third of the cars destined for the Tenderloin were from outside the neighborhood. The greater use of cars by non- Tenderloin residents was even greater when only work trips were analyzed. Employees who work in the Tenderloin, but live elsewhere, were more likely to drive into the Tenderloin for work. The auto mode share for all San Francisco residents with ori- gins in the Tenderloin (35.4% for work trips) was double the automobile mode share of trips made only by Tender- loin residents (17.7%). This difference can be explained by a large number of Tenderloin workers who commuted from outlying neighborhoods by private automobile. The specific characteristics of residents versus nonresidents making trips in the neighborhood were easy to analyze because of the disaggregate nature of the SF- CHAMP outputs, which thus provided a new way of using model results to support planning project work. New Starts SFCTA developed an application of the San Francisco model to the proposed New Central Subway project in downtown San Francisco (5). This is the first application of a tour- based travel demand model in the United States to a major infrastructure project in support of a submis- sion to the FTA for project funding through the New Starts program. To enable the submittal of a New Starts request, software was developed to collapse the microsimulation output of the models for tour and trip mode choice into a format compatible with the FTA SUMMIT program. SUMMIT was then successfully used to summarize and analyze user benefits accruing to the project and to pre- pare an acceptable New Starts submittal. Parallel Processing The initial implementation of the SF- CHAMP model took 36 h to run, which became a major impediment to both further model development and application. The bulk of this time was not in core microsimulation steps but rather in the road and transit skim- building and assignment procedures. The desire to decrease random microsimulation variation (by running multiple itera- tions), combined with the highly granular nature of the skim- building and path- building steps, made obvious the need for a parallel structure instead of the existing top- down model process. SFCTA devised a job control system to allow a model job to be submitted as a transaction, which would then be processed by all available machines as quickly as pos- sible, in parallel. The most difficult aspect of this process was analyzing the dependency tree of model steps to determine which ones could be made parallel and which could not; some steps obviously required that earlier actions be complete before the steps could be made par- allel. Job files were rewritten to unlink the pieces that did not depend on each other. The revised job files were passed to a new dispatcher utility program that could allocate each step to available computers and keep track of the model run progress. The extraordinary time saving of this method was limited only by the amount of hardware available and the granularity of the model steps. In practice, full runs shortened from 36 to 9 h. The goal of an overnight run thus attained, staff added five additional core iterations to reduce error due to microsimulation variability. The model now runs in just under 12 h. REFERENCES 1. Cambridge Systematics, Inc. San Francisco Travel Model Development Executive Summary. San Francisco County Transportation Authority, San Francisco, Calif., June 30, 2001. 2. Cambridge Systematics, Inc. San Francisco Travel Model Development Report on Validation of 1998 Models. San Francisco County Transportation Authority, San Francisco, Calif., May 7, 2001. 3. Cambridge Systematics, Inc. San Francisco Travel Model Development Report on MTC Consistency. San Francisco County Transportation Authority, San Francisco, Calif., May 15, 2001. 4. Castiglione, J., R. Hiatt, T. Chang, and B. Charlton. Appli- cation of Travel Demand Microsimulation Model for Equity Analysis. In Transportation Research Record: Jour- nal of the Transportation Research Board, No. 1977, Trans- portation Research Board of the National Academies, Washington, D.C., 2006, pp. 35â42. 5. Freedman, J., J. Castiglione, and B. Charlton. Analysis of New Starts Project by Using Tour- Based Model of San Francisco, California, In Transportation Research Record: Journal of the Transportation Research Board, No. 1981, Transportation Research Board of the National Academies, Washington, D.C., 2006, pp. 24â33. 29THE SAN FRANCISCO MODEL IN PRACTICE