Expand Configuration and Management Manual (G06.24+)

Tuning
Expand Configuration and Management Manual—523347-008
20-28
Measuring Passthrough Traffic
•
CPU-Busy-Time = 31.20%
•
Process Dispatches = 48.13
•
CPU Dispatches = 51.00
•
CPU Intr-Busy-Time = 13.24%
•
CPU Send-Busy-Time = 4.64%
Using the formula shown above, the adjusted processor use for $PATHF is 39.32%, as
reached from the following formula:
For a complete accounting of Expand line-handler process overhead, the network
control process ($NCP) and Expand manager process ($ZEXP) should also be
measured. $NCP activity is relatively fixed and depends on network size, configuration,
and the quality of the transmission lines. Because the activity of $NCP is not directly
related to the amount of traffic, it is not necessary to account for it when sizing or
tuning a path.
Measuring Passthrough Traffic
Although passthrough traffic is reported in the SYSTEM entity (SENT-FWD and RCVD-
FWD counters), Measure does not directly account for the source and destination of
passthrough traffic when examining a path. An Expand line-handler process only sees
the node to which it is connected. The only way to map the passthrough traffic
accurately is to know the topology of your network and to measure each Expand line-
handler process on each node. This is usually not feasible or necessary for most
tuning efforts.
Setting Measurement Intervals
It is important to set appropriate measurement intervals when using Measure to
characterize Expand network performance. For example, if you want to define the peak
traffic of a path at a critical hour during the day, you should take a sample at short
intervals of perhaps 1-minute during the sample period. To gather information for a
daily traffic profile, you should take a sample every 30 minutes during a 24-hour period.
Regardless of your objectives, you should make sure that you take samples often
enough so that you do not allow Measure’s counters to overflow.
You can simultaneously measure the same Expand entities with more than one interval
using two separate measurement files. An effective management strategy is to sample
network performance at long intervals (possibly every hour on a regular basis) while
simultaneously taking short-interval snapshots during critical periods for a particular
study.
31.20 + ((48.13 / 51.00) * (13.24 - 4.64)) = 39.32
Note. Measuring adds no burden to the measured entity; the counters maintained by the
system are sampled by Measure without affecting the measured process.