Results 1 - 10
of
80,948
Table 2: Amortized processing times (hypercube) Fat-tree, multiple queries Binary tree, single query
"... In PAGE 10: ... In this way (for more details, see Appendix A and [5]), we get a reliable platform for a fair experimental comparison between the multiple query algorithm and the binary{tree single{query algorithm, as well as for a fair comparison between the two versions of our multiple query algorithm based on the hypercube and fat{tree topology respectively. Speci cally, in the rst table16 ( Table2 ), the amortized processing times (Rp, in seconds) as well as the corresponding achieved utilization values (Up17, in percent values) are presented, for both the hypercube{ based multiple{query algorithm and the binary{tree single{query algorithm. We present measurements for varying number of processors (16, 32, 64, 128) and for varying collection size (1/4{WSJ, 1/2{WSJ and full WSJ collection, by appropriately dividing the total number { almost 75000 { of the WSJ documents).... ..."
Table 7: Measurements for the fat-tree algorithm
1996
Cited by 5
Table 7: Measurements for the fat-tree algorithm
Table 2: Results of the random permutations of messages The most striking result is the poor performance of the meshes, even if one takes into account that they have a larger diameter. While the average transfer time of rings and fat trees is relatively close to the optimal transfer time (which is not surprising since random communication patterns tend to balance the load), the average transfer time in the mesh is much worse than the optimal transfer time. Histograms of delays per message in the 4 mesh and 16 fat tree highlight the large variation of delays due to multiple collision in the mesh (see Fig 9)
"... In PAGE 11: ...lock is. This number can be doubled by folding the upper level of switches. Now, for a same number of nodes, one can choose to interconnect a certain number of small blocks or half this number of large blocks. We will see later on simulation ( Table2 ) that the rst solution is the best. Table 1 gives the maximum number of nodes that we can obtain with a 3 level fat-tree topology depending on the redundancy between levels.... In PAGE 12: ...Table2 shows the results of the rst series of simulation. Networks are grouped by topology and arranged by size.... ..."
Table 3.6 shows the relationship between network and I/O errors. In looking down the
Table 7: The I/O costs of Network Operations for various access methods
1997
"... In PAGE 26: ... Page under ows and over ows in the Delete() and Insert() operations are ignored to lter out the e ect of reorganization policies, which are studied separately. Table7 shows the average number of data page accesses for each operation under various methods. The CRR value for each method is also listed in the table.... In PAGE 26: ...Table 7: The I/O costs of Network Operations for various access methods As shown in Table7 , the number of data page accesses during Get-A-successor(), Get-successors() and Delete() operations with CCAM-S is the lowest among all the methods. This is to be expected, since CCAM-S has the highest CRR.... In PAGE 31: ... In general, the I/O cost ranking for all six access methods is the same for all three route length classes, but the I/O gap between the di erent access methods increases as the route length increases. The CRR values for the various access methods are listed in Table7 . As we expected, access methods with a higher CRR have a lower I/O cost for path computation.... ..."
Cited by 41
Table 5: Locality Structure of Full Fat-Trees
1991
Cited by 2
Results 1 - 10
of
80,948