### Table 4 and Table 5 present the performance of CHARMM when atom decomposition and force decomposition algorithms are applied for iteration partitioning, respectively. The execution time in- cludes the energy calculation time and communication time of each processor. The computation time is the average of the computation time of dynamics simulations over processors; the communication time is the average communication time. As predicated in Section 4.2, force decomposition has lower communication overheads than atom decomposition. The results also show that CHARMM scales well 2Estimate (based on model systems due to memory limitations with 1 node) [4]

1995

"... In PAGE 13: ... The irregular distribution generated by weighted RCB achieves good load balance because the weighted version of RCB assigns nearly the same amount of computation load to each processor. Performance Table4 : Performance of Parallel CHARMM (Atom Decomposition)... In PAGE 14: ... The schedule regeneration time in Table 6 gives the total schedules regeneration time for 40 non-bonded list updates. This overhead was included in the computation times in Table4 and Table 5. By Comparing these numbers to those in Table 5, it can be observed that the preprocessing overhead is relatively small compared to the total execution time.... ..."

Cited by 47

### Table 4 and Table 5 present the performance of CHARMM when atom decomposition and force decomposition algorithms are applied for iteration partitioning, respectively. The execution time in- cludes the energy calculation time and communication time of each processor. The computation time is the average of the computation time of dynamics simulations over processors; the communication time is the average communication time. As predicated in Section 4.2, force decomposition has lower communication overheads than atom decomposition. The results also show that CHARMM scales well 2Estimate (based on model systems due to memory limitations with 1 node) [4]

"... In PAGE 13: ... The irregular distribution generated by weighted RCB achieves good load balance because the weighted version of RCB assigns nearly the same amount of computation load to each processor. Performance Table4 : Performance of Parallel CHARMM (Atom Decomposition)... In PAGE 14: ... The schedule regeneration time in Table 6 gives the total schedules regeneration time for 40 non-bonded list updates. This overhead was included in the computation times in Table4 and Table 5. By Comparing these numbers to those in Table 5, it can be observed that the preprocessing overhead is relatively small compared to the total execution time.... ..."

### Table 2. Comparison of pass rates of multipath fading distributions using full band

"... In PAGE 3: ... Tables below show, how different CDFs fit to the data. Table2 depicts all of the measurement environments from LOS to NLOS2 (c.f.... ..."

### TABLE I PREDICTED SCALING OF A GENERIC TILE-BASED SYSTEM

### Table 6: Combining Phase Truncation with Dominance Decomposition: Performance of Large-Scale Voice/Video Integration

in Analysis of Multimedia Traffic Queues with Finite Buffer and Overload Control, Part II: Applications

1992

"... In PAGE 12: ... The deletion for type 2 is again by two symmetric cuts on both sides of the average state, which is 240 0:4 = 96. Table6 lists the approximation results by di erent degrees of truncation. ~ N represents the size of R by the product of the sizes of the two reduced sub-phase-spaces.... In PAGE 12: ... In spite of this sensitivity, the absolute error of the loss rate is found to be bounded by the tail probability of phase truncation quot;. The data on the rst two rows of Table6 illustrate an interesting relationship between phase truncation and dominance decomposition. In particular, the data on the rst row are obtained by applying both techniques, namely, reducing N1 = 21 to ~ N1 = 17 and N2 = 240 to ~ N2 = 1.... ..."

Cited by 21

### TABLE 1. Distribution of Availability of Healthy Foods Scale and Area Characteristics by Study Site Using Data From the Community Survey (n H11005 5186), 2004, InfoUSA 2003, and Census 2000

2006

### Table 1: Average state predictabilities for three topologies

1996

"... In PAGE 16: ... So, for any e ective communication path s, its system state predictability is calculated as SSP (s) = E path(P (V; E)) ? E path(P=s(V=s; E=s)) E path(P (V; E)) : In addition, we must distinguish the e ective communication paths based on the system symmetry so that we can reduce the computational complexity of formula (5). Based on SSP metric, we quantify in Table1 the e ectiveness of the token-chasing algorithm on three kinds of network topologies: a complete topology where each pair of nodes have a direct communication link between them, a sart topology where only the unique central node has links to the other nodes, and a ring topology. Here N is the number of nodes in a topology.... In PAGE 16: ... Here N is the number of nodes in a topology. Comparing the analytical results in Table1 , we can get following conclusions:... ..."

Cited by 1

### Table 3: Income classes for the United States and for Italy US Italy

### Table 4: Phase behavior of programs with visible large scale cycles

1999

"... In PAGE 13: ... The perl graph in figure 6 shows during the initialization that the value prediction miss rate is 15%, but during the steady state it is 21%, as is summarized in table 3. Table4 shows, for the cyclic programs, the difference between various phases of execution for the above listed statistics. The first group of results are the steady state results when the IPC is high, and the second set of results are when the IPC is low.... ..."

Cited by 57