W e compare the performance of seven methods in computing or approximating service levels for nonstationary M t /M/s t queueing systems: an exact method (a Runge-Kutta ordinary-differential-equation solver), the randomization method, a closure (or surrogate-distribution) approximation, a direct infinite-server approximation, a modified-offered-load infinite-server approximation, an effective-arrival-rate approximation, and a lagged stationary approximation. We assume an exhaustive service discipline, where service in progress when a server is scheduled to leave is completed before the server leaves. We used all of the methods to solve the same set of 640 test problems. The randomization method was almost as accurate as the exact method and used about half the computational time. The closure approximation was less accurate, and usually slower, than the randomization method. The two infinite-server-based approximations, the effective-arrival-rate approximation, and the lagged stationary approximation were less accurate but had computation times that were far shorter and less problem-dependent than the other three methods.
An efficient parallel algorithm is reported for determining all bound rovibrational energy levels for the HO 2 molecule for nonzero angular momentum values, Jϭ1, 2, and 3. Performance tests on the CRAY T3D indicate that the algorithm scales almost linearly when up to 128 processors are used. Sustained performance levels of up to 3.8 Gflops have been achieved using 128 processors for J ϭ3. The algorithm uses a direct product discrete variable representation ͑DVR͒ basis and the implicitly restarted Lanczos method ͑IRLM͒ of Sorensen to compute the eigenvalues of the polyatomic Hamiltonian. Since the IRLM is an iterative method, it does not require storage of the full Hamiltonian matrix-it only requires the multiplication of the Hamiltonian matrix by a vector. When the IRLM is combined with a formulation such as DVR, which produces a very sparse matrix, both memory and computation times can be reduced dramatically. This algorithm has the potential to achieve even higher performance levels for larger values of the total angular momentum.
The discrete variable representation (DVR) method has been
modified in three major ways to produce a
more efficient scheme for calculating the rotational−vibrational
energies of van der Waals molecules. First,
the implicitly restarted Lanczos method (IRLM) of Sorensen (SIAM
J. Matrix Anal. Appl.
1992, 13, 357)
is
used to determine the eigenpairs of interest. Second, Chebychev
polynomial preconditioning is applied to
make it easier to project out unwanted eigenfunctions and thus speed up
the convergence of the IRLM. Finally,
a very efficient matrix−vector algorithm is introduced that makes
maximum use of the underlying sparsity
of the DVR Hamiltonian. Calculations for a protypical system
Ar−HCl are reported. For the 20 lowest
Ar−HCl eigensolutions corresponding to angular momentum J
= 1 and using only a single processor of a
Cray YMP, the modified DVR approach is about 6 times faster than the
original DVR method and 14 times
faster than the collocation method of Peet and Yang (J. Chem.
Phys.
1989, 91, 6598). As the total
angular
momentum is increased, the relative performance of the modified DVR
approach improves dramatically.
For instance, with J = 5 the modified method is about
45 times faster than the original DVR and 100 times
faster than the collocation method on a single processor of a Cray YMP.
This modified DVR approach also
runs significantly faster on a range of workstations such as DEC Alpha
and the IBM RS/6000. Application
of this method is also shown to be effective in obtaining an improved
interaction potential for the
A
2Σ+ state
of Ar−HO.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.