### Table 5: The processor-step complexity of many algorithms can be reduced by using fewer processors and assigning many elements to each processor.

1989

"... In PAGE 25: ...rocessor. Such simulation is important for two reasons. First, from a practical point of view, real machines have a xed number of processors but problem sizes vary: we would rather not restrict ourselves to xed, and perhaps small, sized problems. Second, from both a practical and theoretical point of view, by placing multiple elements on each pro- cessor, an algorithm can more e ciently utilize the processors and can greatly reduce the processor-step complexity (see Table5 ). Figure 10 discusses how to simulate the various vector operations discussed in Section 2.... ..."

Cited by 138

### Table 5. Parallel performance of the eigenvalue computation (average time for the linear solver in every step) for P2-elements. Due to the cpu-time-limit, not suf ciently many steps of the block inverse iteration could be performed on level 4 (32 proc.) and level 5 (128 proc.).

1999

Cited by 2

### Table 4: Parallel performance of the eigenvalue computation (average time for the linear solver in every step) on IBM RS/6000 for P2-elements. Due to the cpu-time-limit, not su ciently many steps of the block inverse iteration could be performed on level 4 (32 proc.) and level 5 (128 proc.).

### Table 7: This table illustrates the steps used in the BGP de- cision algorithm for route updates. Each entry illustrates how many times a particular step resulted in a tie-breaking event.

2006

"... In PAGE 6: ...Which steps in the BGP decision algorithm are most im- portant? Table7 quantifies the tie-breaking steps in the BGP decision making algorithm. We expected MED and Hot Potato to play a larger role in the algorithm, based on previous work [13, 12].... ..."

Cited by 1

### Table 3: The relative di erence ( ^ V2 ? ^ V1)= ^ V1 in percentage for each sample pair ( ^ V1; ^ V2), broken down by volatility band and portfolio subset. ^ V2 is computed with twice as many time steps as ^ V1. No measurements are available for max = 100%

1998

"... In PAGE 23: ...n series 1 and 2, respectively (the subscripts are dropped for convenience, i.e. ^ V = ^ V 0 0 ). As shown in Table3 , convergence is better for narrow volatility bands. For max = 20% we nd stability in the rst two leading digits; for max = 40 and 60%, a large portion of the results disagree on the second leading digit.... ..."

### Table 2: BLAS routines called in GMRES 4 An example of use We give below an example of use of the FGMRES driver. Here the preconditioner is the GMRES method implemented as in [4]. Note that, in this example, we have chosen not to allocate extra memory for the preconditioner: when the right preconditioner is needed for FGMRES, we compute how many steps of GMRES are possible with the part of the workspace which is still free. The inner GMRES can be itself preconditioned: in this example we propose a Jacobi (left) preconditioner.

1998

Cited by 5

### Table 2. 5-step characteristic.

"... In PAGE 14: ... Therefore good characteristics for these steps should have low Hamming weights through as many steps as possible. Consider rst the 5-step characteristic of Table2 . The rst four steps... ..."

### Table 1: Best polling strategies for maximum delay=6. The table shows how many locations should be polled at each time step for various interpolling interval values.

1998

"... In PAGE 8: ... Experiments were performed using a maximum delay tolerance of 6 for larger inter- polling intervals (2 to 801) to investigate the limiting behavior of the strategies and cost function values. The results are shown in Figure 6 with the group sizes in Table1 . Notice that as the interpolling interval becomes very large the optimum strategy approaches the classical \blanket polling quot; strategy where all locations are polled simultaneously in the rst step.... ..."

Cited by 12

### Table 2: Best polling strategies for maximum delay=5. The table shows how many locations should be polled at each time step as the delay factor in the cost function increases.

1998

Cited by 12

### Table 1: Best polling strategies for maximum delay=6. The table shows how many locations should be polled at each time step for various interpolling interval values. II = Interpolling Interval

1996

Cited by 3