Information Technology Reference
In-Depth Information
Algorithm 2: Update Flow Forecast
f
f
BW
¨
¨•
¨-
¨
¨•
BW
old
f
min
min
BW
TT T
TT
T
REM
min
f
f
elap
NOW
start
f
start
NOW
f
end
_
SIZE
f
¨
=
REM
_
SIZE
f
old
f
f
f
f
REM
_
SIZE
REM
_
SIZE
-
(
T
-
BW
)
old
elap
old
min
f
for each
do
BW
Œ
<
BW
uv
,
f
f
{
if
BW
BW
then
BW
=
BW
}
uv
,
min
min
uv
,
f
REM
_
S
IZE
f
f
T
=
T
+
dur
f
BW
TTT
min
f
f
f
¨+
end
start
dur
15.8 Performance Comparison
In this section we quantify the performance improvements gained from
modeling networking trafi c as
l ows
instead of packets. Using our candi-
date implementation of a l ow model for the GridSim toolkit (described
previously) we perform some numerical comparisons of specii c scenarios.
In each scenario we compare the existing NetPacket implementation and
the new l ow-level FlowPacket implementation described in this chapter.
All tests are run using a Macbook with a 2 GHZ Intel Core 2 Duo and 2 GB
of RAM. Each data point represents the average of 30 runs.
The i rst scenario is a classic dog-bone topology similar to that depicted
in
Figures 15.2
and
15.3
.
NetUser1/FlowUser1 and NetUser2/FlowUser2
each send three identically sized i les to Resource6 and Resource5, respec-
tively. The size of the i les is varied from 0.5 to 500 MB. The links between
the users and the i rst router are rated at 10 MB/s. The link between the
two routers has a capacity of 1.5 MB/s, and is clearly the bottleneck link.
The link between the second router and the resources is 10 MB/s. The i les
transfers are initiated in 10-second intervals. The latencies from the users
to the router, between the routers, and from the router to the resources are
45, 25, and 30 milliseconds, respectively.
In
Figure 15.5
we see a comparison of the CPU time taken for the simu-
lation to execute when utilizing either the NetPacket or FlowPacket net-
working stack. The size of the i les sent by each user is varied from 0.5 to
Search WWH ::
Custom Search