The objective is to maximize this determinant (if there is no correlation at all, it is equal to 1 and the design is orthogonal).įigure 4.8. This criterion is calculated using the determinant of the parameter correlation matrix. This means that they are as uncorrelated as possible (if there is a correlation between two parameters A and B, it is difficult to separate the effect of A from the effect of B). Independence: it is desirable for the parameters to be as orthogonal as possible. The objective is to maximize this distance − This criterion is calculated using the distance between the closest two points on the design. Space-filling: the trials of the LHS DOE should fill the input domain as much as possible. The main characteristics of LHS designs are: − This design is not unique and the space-filling and independence criteria are relative. In this case, a very high number of designs are generated and the design which contains the best space-filling and independence criteria is selected. LHS designs are particularly suitable for numerical designs of experiments. The objective is to maximize this determinant (if there is no correlation at all, it is equal to 1 and the design is orthogonal). This means that they are as uncorrelated as possible (if there is correlation between two parameters A and B, it is difficult to separate the effect of A from that of B). The objective is to maximize this distance – Space-filling: it is desirable for the design’s test to fill the input domain as much as possible. The main properties expected for LHS designs are: – We evaluate their throughput using simulations, and compare it to that of the priority scheme.* Values in parenthesis are design values. We also consider two deflection routing schemes, called the simple nonwasting deflection and the priority nonwasting deflection schemes. We find that little buffer space (between one and three packets per link) is necessary to achieve throughput close to that of the infinite buffer case. The results obtained are approximate, but very accurate as simulations indicate, and are given in particularly interesting forms. We evaluate the throughput of both the unbuffered and the buffered version of these schemes for random multiple node-to-node communications. We evaluate their throughput using simulations, and compare it to that of the priority scheme.ĪB - We consider two different hypercube routing schemes, which we call the simple and the priority schemes. N2 - We consider two different hypercube routing schemes, which we call the simple and the priority schemes. Bertsekas is with the Laboratory for Information and Decision Systems, Massachusetts Institute of Technology, Cambridge, MA 02139 USA. Varvarigos is with the Department of Electrical and Computer Engineering, University of Califomia, Santa Barbara, CA 93106 USA. This work was supported by the NSF under Grants NSF-DDM-8903385 and NSF-RIA-08930554, and by the ARO under Grants DAAL03-86-K-0171 and DAAL03-92-G-0309. Manuscript received Decemrevised Septemapproved by IEEWACM TRANSACTIONSO N NETWORKINEGdi tor I. T1 - Performance of Hypercube Routing Schemes With or Without Buffering
0 kommentar(er)
