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.
dom number series to determine which possibility is cho- sen for that record. The new model is implemented with three global feed- back loops for consistency between highway travel times that are both used as inputs to, and as forecast outputs of, the model. The main model application package is Cube, with TP+ being used for network management, assignment, external, and commercial vehicle models and other pro- cessing. The core tour- based choice models described above are written in Java with access to the TP+ skims. The custom programs are designed to take advantage of the numerous opportunities for parallel processing in the model chain, multi- threading of tasks, and to readily accommodate the addition of computers in the distribu- tive processing framework to optimize processing. After the networks and initial skims are generated in TP+ and all input files are created for a particular sce- nario, the custom Java programs are executed to imple- ment the tour- based microsimulation models. A pre- assignment processor step aggregates the microsimu- lation results and integrates the commercial and external models to produce standard TP+ trip tables for four time periods. After the final trip tables are generated, vehicles are assigned with a multi- class (SOV, HOV, medium truck, and heavy truck) equilibrium assignment utilizing 21 volume delay functions by facility and area type for each of four time periods (a.m., midday, p.m., and night). Transit assignments are also performed in TP+ for the a.m. and midday time periods, with standard reports generated to support analysis and evaluation of the alter- natives tested (Anderson et al. 2003). HARDWARE CONFIGURATIONS There are three operational systems that can run the MORPC travel forecasting model. The two systems that are currently installed at MORPC are the topic of this paper. The initial system was built in December 2004 with one server computer and three worker computers. The specifications for the computers are below. ⢠Server â Dual 64-bit Xeon 3.6 GHz 1MB L2 800MHz FSB Processors â 4 GB PC3200 ECC Registered DDR Memory â 4 - 36GB SCSI 15K U320 RAID-5 Array â Dual Gigabit network interface cards â Windows 2000 Server Operating System (OS) ⢠Worker (3â4) â Dual Xeon 3.06 GHz 512KB L2 533MHz FSB â 2 GB PC2100 ECC Registered DDR Memory â 120 GB IDE HDD â Gigabit network interface card â Linux 32-bit OS ⢠Networking Specifications â 5 port 10/100/1000 Gigabit network switch â CAT6 Ethernet cable The workers are directly networked and are isolated from the general MORPC network to make them less susceptible to viruses. The workers are not running anti- virus software every time a file is accessed, unlike the rest of the MORPC network; it was found that running anti- virus software imposed a 15% penalty on the run time. In December 2005, a fourth worker was added to the cluster. This first system is running 32-bit OSs and Java. The second system was purchased by the Central Ohio Transit Authority (COTA) in support of its North Corridor Transit Project DEIS. This cluster consists of one server and four workers all running 64-bit Windows and Java. The specifications for this cluster are below. ⢠Server â Dual 64-bit AMD Opteron 2.2 GHz 1MB L2 Cache Processors â 4 GB PC3200 ECC Registered DDR Memory â 4 - 73GB SCSI U320 10K RPM RAID-5 Array â Dual Gigabit network interface cards â Windows 2003 Server 64-bit OS ⢠Worker (4) â Dual 64-bit AMD Opteron 2.2 GHz 1MB L2 Cache Processors â 4 GB PC3200 ECC Registered DDR Memory â 160 GB SATA NCQ HDD â Dual Gigabit network interface card â Windows XP Professional 64-bit OS MODEL RUNNING TIMES Table 1 shows the running times of the MORPC Travel Forecasting Model for 2000 and 2030 on the various computer systems. MORPC 3 is the MORPC system with three Linux workers running 32-bit OSs, MORPC 4 is the MORPC system with four Linux Workers run- ning 32-bit OSs, and COTA is the COTA system with four Windows XP Workers running 64-bit OSs. The 2000 model has 1.5 million synthetic people making 2 million tours; 2030 has 2 million synthetic people mak- ing 3 million tours. All runs include only two transit modes (local and express bus). The core model running time does not include the time to generate the four- period highway networks, two- period transit networks includ- ing support links, or the time to generate the initial travel skims, which is similar to the time to generate the travel skims during the model run. Overall, the time for these excluded tasks is about 2 h of running time on a 32-bit Windows computer. 182 INNOVATIONS IN TRAVEL DEMAND MODELING, VOLUME 2