• 沒有找到結果。

Evaluation under Different Types of Fork-Join Structures

Chapter 6 Performance Evaluation

6.2 Evaluation under Different Types of Fork-Join Structures

In this section we classify the workflows of fork-join structure into four types and evaluate the proposed clustering approaches under each type of workflows accordingly.

Figure 6.1 is an example of the first type of workflows where the edges connecting to the fork node have larger communication costs than the edges pointing to the join node. Figure 6.2 and Figure 6.3 show the performance evaluation of the proposed clustering approaches with the first type of workflows. The performance was measured by the reduction of the average makespan of all workflows, compared to the PCH approach [10]. The experimental results indicate that the proposed CPCH, SLECPCH and SCCCPCH schemes outperforms PCH significantly, up to 3% performance improvement.

Figure

Figure 6.2: Performance with the first type of workflows of single

0 1 2 3 4 5

CPCH 4.18

Performance improvement in makespan

Figure 6.1: First type of workflows

Performance with the first type of workflows of single-level fork

CPCH SLECPCH SCCCPCH 4.18

1.75

2.53

Single-level fork-join

level fork-join

Figure 6.3: Performance with the first type of workf

Figure 6.4 is an example of the second type of workflows where the edges connecting to the join node have larger communication costs than the edges starting from the fork node.

Figure 6.5 and Figure 6.6 show the performance evaluation of the approaches with the second type of workflows.

proposed CPCH, SLECPCH and SCCCPCH schemes outperforms PCH significantly, up to 13% performance improvement

of workflows, compared to the first type of workflows.

Figure

Performance with the first type of workflows of double-level fork

Figure 6.4 is an example of the second type of workflows where the edges connecting to the join node have larger communication costs than the edges starting from the fork node.

6.5 and Figure 6.6 show the performance evaluation of the proposed approaches with the second type of workflows. The experimental results

SLECPCH and SCCCPCH schemes outperforms PCH significantly, up to ce improvement. A larger performance improvement is achieved with this type of workflows, compared to the first type of workflows.

Figure 6.4: Second type of workflows

CPCH SLECPCH SCCCPCH Figure 6.4 is an example of the second type of workflows where the edges connecting to the join node have larger communication costs than the edges starting from the fork node.

proposed clustering results show that the SLECPCH and SCCCPCH schemes outperforms PCH significantly, up to . A larger performance improvement is achieved with this type

Figure 6.5: Performance

Figure 6.6: Performance with the second type of workflows of double

Figure 6.7 shows an example of the third type of workflows where the edges on the upper path between the fork and join nodes have larger communication costs than the edges on the lower path. Figure 6.8 and Figure 6.9 show the performance evaluation of the

0

with the second type of workflows of single

-Performance with the second type of workflows of

double-Figure 6.7 shows an example of the third type of workflows where the edges on the upper path between the fork and join nodes have larger communication costs than the edges

6.8 and Figure 6.9 show the performance evaluation of the

CPCH SLECPCH SCCCPCH upper path between the fork and join nodes have larger communication costs than the edges 6.8 and Figure 6.9 show the performance evaluation of the proposed

clustering approaches with the third type of workflows.

the proposed CPCH, SLECPCH and SCCCPCH schemes outperforms PCH significantly, up to 12% performance improvement

superiority of the three clustering

Figure

Figure 6.8: Performance with the third type of

0

clustering approaches with the third type of workflows. The experimental

SLECPCH and SCCCPCH schemes outperforms PCH significantly, up performance improvement. Unlike the previous two types of workfl

superiority of the three clustering approaching remains the same in these two figures.

Figure 6.7: Third type of workflows

Performance with the third type of workflows of single-level fork

CPCH SLECPCH SCCCPCH

13.9 14.12

17.97

Single-level fork-join

results indicate that SLECPCH and SCCCPCH schemes outperforms PCH significantly, up . Unlike the previous two types of workflows, the relative

s the same in these two figures.

level fork-join

Figure 6.9: Performance with the third type of workflows of double

Finally, Figure 6.10 is an example of the fourth type of workflows where the edges on the lower path between the fork and join nodes have larger communication costs than the edges on the upper path. Figure

proposed clustering approaches with the fourth type of workflows.

indicate that the proposed CPCH,

significantly, up to 11% performance improvement clustering approaching remain

Performance with the third type of workflows of double-level fork

Finally, Figure 6.10 is an example of the fourth type of workflows where the edges on een the fork and join nodes have larger communication costs than the Figure 6.11 and Figure 6.12 show the performance evaluation of the clustering approaches with the fourth type of workflows. The experimental

that the proposed CPCH, SLECPCH and SCCCPCH schemes outperforms PCH performance improvement. Again, the relative superiority of the three remains the same in these two figures.

Figure 6.10: Fourth type of workflows

CPCH SLECPCH SCCCPCH

25.85 26.02

29.24

Double -level fork-join

level fork-join Finally, Figure 6.10 is an example of the fourth type of workflows where the edges on

een the fork and join nodes have larger communication costs than the 6.11 and Figure 6.12 show the performance evaluation of the experimental results SLECPCH and SCCCPCH schemes outperforms PCH . Again, the relative superiority of the three

Figure 6.11: Performance with the fourth type of workflows of single

Figure 6.12: Performance with the fourth type of

6.3 Evaluation under a Mixture of Various Fork

相關文件