|
A merit for user to utilize web-based on-line scheme is that it can be used by any network-enabled computer system
without restriction of space and/or time. In conjunction with the accuracy of on-line module,
another important considerations on on-line scheme are computational time and
memory usage on server-side. From user's viewpoint, it is obvious that too much computational time
will reduce a merit of this on-line simulation, thus it is required to shorten computational time.
In addition, it is not desirable for server system to go over the limitation of its physical memory (RAM)
caused by numerous simulating locations being in general required much more server memory as well as very long computational time.
Comparing four schemes to computational time and memory usage,
Wittig and Sinha (1975), Di Paola and Gullo (2001) and Shinozuka and Deodatis (1991) schemes have more merits in both categories than Deodatis (1996) scheme.
Since Wittig and Sinha (1975) scheme utilizes discrete frequency function by introducing FFT and Cholesky decomposition,
the scheme is fast and requires relatively small memory usage.
In case of Di Paola and Gullo (2001) scheme, it utilizes Schur decomposition which is required eigenvalue analysis similar to POD approach
(e.g., Kareem 1978; Kareem 1999; Chen and Kareem 2005; Chen and Letchford 2005; Ding et al. 2006),
while by using only first eigenvalue and eigenvector, required memory capacity can be dramatically reduced.
The main reason that Deodatis (1996) scheme scores low in both categories is that the scheme is designed to guarantee
'ergodic process' in single simulation being different from other three schemes used in this module. To accomplish 'ergodic' simulation,
Deodatis (1996) utilized 'double index frequency' concept which requires more memory usage than other two schemes, which resulted in relatively large memory usage and slow computational time, especially when the number of simulation points are quite large.
On the other hand, Shinozuka and Deodatis (1991) scheme, so-called conventional spectral representation method, is essentially similar to Deodatis (1991) scheme, however, it does not guarantee ergodicity in simulated time series.
Accordingly, NOWS has utilized different limitations depending on schemes, especially limiting number of simulated locations per scheme, in view of computational time/memory usage and each scheme's characteristics.
It is important to note that the server has adopted first-come-first-serve approach
to maximize its performance in computation. Accordingly, if there exist preceding user(s) who already queued NOWS execution(s),
the next user(s) in line would have more processing time to get simulation results than actual computational time as single user.
Thus, expected processing time may vary with the number of users waiting in a queue.
|
