NYC-WTC_000187725

NYC 9/11 Public Portal Document

133 Table 5.5 Computational Time on the CRAY-2 Supercomputer for Selected Policy Runs (in Seconds)

Base Policy Policy Policy Policy run (1) (2) (3) (4)

Reading data and 112.83 113.021 113.14 112.97 1,247.9 initial comps. Housing equil. 3.19 6.37 9.77 6.31 6.30 (itera­ (1) (2) (3) (2) (2) tions) Commercial equil. and 28.29 28.61 28.47 28.77 29.09 output computation Policy (1): 5% decrease in auto commuting and shopping time. Policy (2); 10% decrease in auto commuting and shopping time. Policy (3): 5% decrease in reported subway commuting time. Policy (4): decrease the subway waiting time from 5 minutes to 2.5 minutes (input data is network based times). The convergence tolerance is 0.5 % of the equilibrium rents.

When rents are perturbed within a 50% b_and of the equilibrium values, then more than 99 iterations are required to achieve convergence. Similarly starting with uniformly zero rents requires a large number of iterations. In practice, it would be very difficult to find conditions under which the equilibrium rents would be that different from the initially known equilibrium rents. Hence it should be relatively easy to keep the overall computational cost of any one run under 456 seconds.

Computational Experience on the SUN Workstation Because supercomputers are still relatively inaccessible, the empirically calibrated model was also extensively tested on a SUN Sparc Station which is a desktop computer with capabilities equal to or better than many mainframes. A SUN Sparc Station and associated equipment can be purchased for about $12,000.

CITY_2IMC10 0_0 0 6 6 6236 4

NYC-WTC_000187864