### Table 10: Data passing costs estimated according to net network transmission rate and memory and L2-cache copy bandwidths are consistent with the actual values. We used data from Tables 5, 6, 7, and 8 to verify the scaling model. Table 10 shows the veri cation of (1), (3), and (4) for each platform. Table 11 shows the veri cation of (3), (4), and (5) across platforms. Agreement between estimated and actual scaling was quite good for the Gateway P5-90, which has the same architecture as the base case. In the AlphaStation, CPU- dominated ratios had geometric means consistent with the model but variances that were much higher than those of the Gateway P5-90. This could be expected, given that the AlphaStation has a substantially di erent architecture.

1996

Cited by 2

### Table 10: Data passing costs estimated according to net network transmission rate and memory and L2-cache copy bandwidths are consistent with the actual values. We used data from Tables 5, 6, 7, and 8 to verify the scaling model. Table 10 shows the veri cation of (1), (3), and (4) for each platform. Table 11 shows the veri cation of (3), (4), and (5) across platforms. Agreement between estimated and actual scaling was quite good for the Gateway P5-90, which has the same architecture as the base case. In the AlphaStation, CPU- dominated ratios had geometric means consistent with the model but variances that were much higher than those of the Gateway P5-90. This could be expected, given that the AlphaStation has a substantially di erent architecture.

1996

Cited by 2

### Table 1: Computational results (times in seconds are obtained on a DEC AlphaStation 500/400Mhz).

### Table 2: Computational results on di erent classes of SDP for Algorithm IPC. Ten random instances are considered for each class. The computations were done on a DEC AlphaStation/500 (333MHz). The number X Z above is the smallest number such that relative duality gap X Z=(1 + jC Xj) is less than 10?6 and the infeasibility measure is less than the relative duality gap.

1999

"... In PAGE 36: ...5 11.5 Table 3: Same as Table2 , but for the homogenous predictor-corrector algorithm, Algorithm HPC. The duality gap X Z above is the smallest number such that the relative duality gap X Z=(1 + jC Xj) is less than 10?6 and the infeasibility measure is at most 5 times more than the relative duality gap.... ..."

Cited by 95

### Table 2: Computational results on di erent classes of SDP for Algorithm IPC. Ten random instances are considered for each class. The computations were done on a DEC AlphaStation/500 (333MHz). The number X Z above is the smallest number such that relative duality gap X Z=(1 + jC Xj) is less than 10?6 and the infeasibility measure is less than the relative duality gap.

1999

"... In PAGE 34: ...5 11.5 Table 3: Same as Table2 , but for the homogenous predictor-corrector algorithm, Algorithm HPC. The duality gap X Z above is the smallest number such that the relative duality gap X Z=(1 + jC Xj) is less than 10?6 and the infeasibility measure is at most 5 times more than the relative duality gap.... ..."

Cited by 95

### Table 2: Computational results on di erent classes of SDP for Algorithm IPC. Ten random instances are considered for each class. The computations were done on a DEC AlphaStation/500 (333MHz). The number X Z above is the smallest number such that relative duality gap X Z=(1 + jC Xj) is less than 10?6 and the infeasibility measure is less than the relative duality gap.

1999

"... In PAGE 29: ...5 11.5 Table 3: Same as Table2 , but for the homogenous predictor-corrector algorithm, Algorithm HPC. The duality gap X Z above is the smallest number such that the relative duality gap X Z=(1 + jC Xj) is less than 10?6 and the infeasibility measure is at most 5 times more than the relative duality gap.... ..."

Cited by 95

### Table 1: Total search times of D1D3CSD3CR on the circuit formulas. C-literals whose clause form contains no more literals than the cutoff value is asserted as a clause. Times are CPU seconds on a DEC AlphaStation 200 (233MHz 21064).

2000

"... In PAGE 6: ... A cutoff value of k means that lemmas whose length is k or less are always asserted globally. Table1 shows the results of running Modoc on the circuit formulas. CBCPD8D3BF, which is one of the SAT testers we will be comparing D1D3CSD3CR against in Section 4.... In PAGE 6: ... Figure 3 shows how the total search time changes as different parameters are used for b and c. The formula set is the same used in Table1 . It can be seen that there is a groove from B4bBNcB5 BP B40BM5BN3BM5B5 to B4bBNcB5 BP B45BM0BN2BM0B5.... ..."

Cited by 1

### Table 5: Characteristics of the computers used in the experiments. The integer rating used for the Micron P166 is the listed SPECint95 of the Dell XPS 166. The rating taken as upper bound for the Gateway P5-90 is the listed SPECint95 of the Dell XPS 90, which has a bigger and faster L2-cache. The rating taken as upper bound for the AlphaStation is its listed SPECint base95 because the version of NetBSD used on it could not be compiled with optimizations. The cache and memory bandwidths listed are the peak values we observed using a bcopy benchmark at user level.

1996

Cited by 53

### Table 5: Characteristics of the computers used in the experiments. The integer rating used for the Micron P166 is the listed SPECint95 of the Dell XPS 166. The rating taken as upper bound for the Gateway P5-90 is the listed SPECint95 of the Dell XPS 90, which has a bigger and faster L2-cache. The rating taken as upper bound for the AlphaStation is its listed SPECint base95 because the version of NetBSD used on it could not be compiled with optimizations. The cache and memory bandwidths listed are the peak values we observed using a bcopy benchmark at user level.

1996

Cited by 53

### Table 5: Characteristics of the computers used in the experiments. The integer rating used for the Micron P166 is the listed SPECint95 of the Dell XPS 166. The rating taken as upper bound for the Gateway P5-90 is the listed SPECint95 of the Dell XPS 90, which has a bigger and faster L2-cache. The rating taken as upper bound for the AlphaStation is its listed SPECint base95 because the version of NetBSD used on it could not be compiled with optimizations. The cache and memory bandwidths listed are the peak values we observed using a bcopy benchmark at user level.

1996

Cited by 53