It is also important to compare the differential effects of each TPCM on knowledge sharing behaviors.
H3: Compared with a risk aversion TPCM, an accelerated processing TPCM exhibited by supplier personnel will have a stronger negative influence on the following knowledge sharing behaviors by a boundary spanner in a buying firm:
Interactions Between Relationship Magnitude and TPCM
While it is expected that an accelerated processing TPCM has a stronger negative main effect on knowledge sharing behaviors, we anticipate that a risk aversion TPCM has a stronger negative moderating effect on the association between relationship magnitude and knowledge sharing behaviors.
H5: Compared with an accelerated processing TPCM, a risk aversion TPCM exhibited by supplier personnel has a stronger negative moderating effect in high-magnitude relationships on the following knowledge sharing behaviors by a boundary spanner in a buying firm:
The two factors manipulated are TPCM and relationship magnitude.
For a 4KB cache, TPCM improves the miss rate by 6% to 29% over PH, with a geometric mean of 13%.
The improvement of TPCM over PH is greater for larger caches, because there are more opportunities for avoiding mapping conflicts.
A code-placement algorithm that is less effective than TPCM tends to leave more mapping conflicts.
In all cases, the difference between the PH and TPCM placements is less pronounced for the two-way associative cache than for the direct-mapped cache.
Associative Benchmark Placement Direct-mapped two-way four-way gcc TPCM 4.70 4.71 PH 5.40 4.92 Random 5.61 5.01 4.83 go TPCM 3.37 3.44 PH 3.98 3.63 Random 4.51 3.77 3.54 ghostscript TPCM 2.15 2.06 PH 2.87 2.27 Random 4.33 3.17 2.38 latex TPCM 1.88 1.72 PH 2.27 1.83 Random 3.21 2.28 1.72 perl TPCM 4.08 3.18 PH 5.17 3.41 Random 5.48 3.57 2.44 porky TPCM 2.15 2.04 PH 3.89 2.84 Random 4.67 3.32 2.50 vortex TPCM 3.78 3.90 PH 5.01 4.48 Random 6.57 5.15 4.46 For the perl benchmark, there is a significant difference between the direct-mapped and two-way associative caches, for both placement algorithms.
Finally, the porky and vortex benchmarks have a sufficiently large amount of code and interleaving of execution so that even with the two-way associative cache, the TPCM placement is still significantly better than the PH placement.