Image Processing Reference
In-Depth Information
A theoretical understanding of the network choices is a clear prerequisite to an NCS deployment
decision. he understanding should extend beyond the network protocol performance to the entire
system performance, and an important component of that understanding is the clear definition of
performance metrics that are important to the particular application environment. An example of
metrics that may be important in automotive manufacturing environment can be found in Ref. [],
which is the output of an industrial network performance workshop; in this case, the top ive metrics
were “node performance,” “ease of use and diagnostics tools,” “cost of security/technology,” “capability
for prioritization,” and “time synchronization.” he exercise of this workshop revealed a few impor-
tant aspects of the metric selection process, which are important to the analytical application of these
metrics in NCS decision making. he first is that many of the metrics such as ease-of-use or cost of
complexity may be difficult to quantify, but these metrics must be incorporated into the quantified
decision-making process (i.e., these metrics should not be avoided). Otentimes, a qualitative metric
can be broken down into quantitative sub-metrics (e.g., ease-of-use into average time between system
crashes plus average time to diagnose a problem) so that it can be compared with other factors; the
challenge is understanding the appropriate weighting for these quantifications. The second impor-
tant aspect is that metrics can often overlap significantly. Care should be taken to break these metrics
down into subcomponents so that a nonoverlapping set of subcomponents can be determined. his
will avoid any “double counting” in the decision-making process.
23.4.2 Experimental Perspective
Usually, the theoretical knowledge that can be applied to NCS analysis must be augmented with
experimental knowledge to complete the performance information base. This need for an experi-
mental perspective is especially true when assessing the contribution of end node performance to
the system performance analysis. Node performance should be evaluated in all systems because, as
noted in Section ., node performance has been shown to dominate over network performance
in many manufacturing system environments. his is especially true in higher-level switched Ether-
net systems, where the emphasis is on providing services such as security and self-typing on top of
the data delivery mechanism [,]. These services provide overhead that contributes to network
congestion; in smaller networks, where the node count does not number in the thousands, the degra-
dation in individual node performance due to overhead factors such as these dominate. For example,
Figure . illustrates that, in a particular two-node Ethernet network, node delay accounted for
over % of the entire end-to-end delay []. Figure . (Section ..) gives an example of the
impact of protocol overhead, in this case VPN to support security.
With respect to both network and node performance, specific network overhead factors that
should be considered experimentally include self-typing or higher-level protocols for enabling con-
trol systems such as OPC (discussed in Section ..) and XML,
∗
protocols for security such as VPN,
and choices of Ethernet transport layer protocols such as TCP/IP vs. UDP. he impact of these factors
is explored in detail in Ref. [].
The wireless NCS application environment demands an additional level of experimental data to
provide detailed information about the impact of the medium on NCS performance. In wired sys-
tems, the “wire” is generally considered to be a medium that can guarantee end-to-end delivery,
provided that the protocol on top of it handles issues of congestion and performance (which is
true for most industrial network protocols). There are number of factors in the various wireless
manufacturing environments that can severely compromise wireless NCS performance. For example,
∗
eXtensible Markup Language (XML) is a self-typing language often used to support communications among dissimilar
components on an industrial Ethernet network.
