Information Technology Reference
In-Depth Information
110
NetPacket
FlowPacket
100
90
80
70
60
50
40
30
20
10
5
10
15
20
# of Gridlets
25
30
35
40
FIGURE 15.8
Comparison of peak memory usage for Gridlet simulation run.
only sees a nominal increase in simulation time as the number of Gridlets
sent by each user increases.
In Figure 15.8 we see an examination of the peak memory usage of the
NetPacket and FlowPacket implementations for GridSim, running the
Gridlet scenario described previously. As the number of Gridlets sent by
each user increases, we see a signii cant increase in memory utilization by
the NetPacket implementation after an initial l at response. Conversely,
the FlowPacket sees only negligible increases in memory utilization as the
number of Gridlets increases.
15.9 Conclusion
We have explored the improvements in accuracy and reduction in complex-
ity that can be achieved by utilizing l ow-based networking in grid simula-
tion, instead of packet-based networking, when attempting to model real
TCP/IP networks. From the results obtained it is clear that signii cant
improvements, in the order of many magnitudes, can be made in terms of
speed and scalability when executing complex grid simulations involving
large data transfers and large numbers of grid job submissions. It is clear that
practitioners will not be limited in the size or complexity of the scenarios
they wish to model, allowing them to simulate complex computational grid
Search WWH ::
Custom Search