O'Reilly logo

Mobile Agents by Wilhelm R. Rossak, Peter Braun

Stay ahead with the world's most comprehensive technology and business learning platform.

With Safari, you learn the way you learn best. Get unlimited access to videos, live online training, learning paths, books, tutorials, and more.

Start Free Trial

No credit card required

298 Chapter 8 Evaluation
maximum heap size of 200 MB. The stack size was set to 512 KB. All of the
computers were fully dedicated during the experiments, as were all of the
computer systems.
For most measurements we used the local-area network (LAN) in our
department at the University of Jena, which is a Fast-Ethernet network with
a bandwidth of 100 Mbit/sec where computers are connected via a single
router. Some measurements were done using a fully dedicated Ethernet net-
work with a bandwidth of 100 Mbit/sec and 10 Mbit/sec connected via a
switch.
Measurements of migrationtimesto the computers inWeimar, Darmstadt,
and Irvine were done using our standard Internet connection, which is a
155 Mbit/sec uplink to the German GigaBit Research Network(G-Win), which
has a theoretical bandwidth of 2.5 Gbit/sec. The University of Weimar is also
connected to G-Win using a 155 Mbit/sec uplink. The quality of the network
connections at Darmstadt and Irvine could not be determined.
8.3 Results of the Basic Experiments
8.3.1 Transmission Time with Regard to Code Size and Network Quality
In the first experiment we examined the time for a ping-pong migration of
a single agent with different sizes, through different networks. The agent
was created on tiffany and had to migrate to and from one other agency.
We compared the migration time for the following code sizes: 1685, 4185,
6685, 11,685, 22,685, and 51,685 bytes. The agent’s state is negligible in this
experiment, because it was smaller than 100 bytes. The agent was trans-
mitted using the PushToNext strategy without enabling the code cache. All
migrations were repeated 1000 times.
Figure 8.1 shows the migration times for all high-bandwidth connections.
The destination agency was started on melina. The graph also shows a mea-
surement when the sender agency as well as the receiver agency were located
on the same computer (tiffany) and the agent migrated using the local loop
without using the network.
The best migration performance was achieved using the 100 Mbit/sec net-
work via a switch, where the smallest agent (1685 byte) only needed 23 ms for
a single migration. The migration time only increased slightly, to 34 ms, for
the largest agent (51,685 bytes). Migration using the 100 Mbit/sec network via
a router was only a few milliseconds slower: 25 ms for the smallest agent and
8.3 Results of the Basic Experiments 299
200
150
100
50
0
Time for a ping-pong migration [ms]
10 MBit/sec (switch)
Localhost
100 MBit/sec (router)
100 MBit/sec (switch)
0 10000 20000 30000 40000 50000
Code size [byte]
Figure 8.1 Time for a ping-pong migration between computers tiffany and melina, using
different high-bandwidth networks. The localhost measurement was done on computer
tiffany.
35.5 ms for the largest agent. The measurement using the internal network
loop of the operating system was, surprisingly, slower than both measure-
ments using a 100 Mbit/sec network. Here, one migration takes 28 ms for
the smallest agent and 38 ms for the largest agent. The only explanation
we have found for this so far is that the higher computational load of exe-
cuting two parallel agencies on a single computer. Migrating agents using a
10 Mbit/sec network is noticeably slower than the other network types. The
smallest agent needs about 37 ms for a single migration, and the largest agent
needs about 98 ms.
As can be seen from Figure 8.1, migration time is positively correlated
with the code size of the agent, and the 95% confidence intervals show that,
measured migration times are not significantly different when transmitting
small agents (fewer than 11,685 bytes) using fast networks. This gives hints
for the construction of an optimization strategy: It is not always worthwhile
to reduce a 10 KB agent to 5 KB, because the difference generally cannot be

With Safari, you learn the way you learn best. Get unlimited access to videos, live online training, learning paths, books, interactive tutorials, and more.

Start Free Trial

No credit card required