We need to start 100 SIPp processes using different IP addresses like

sipp -i ${binding_ip} \
  -sf sipp/3pcc-C-A.xml \
  -inf sipp/call-data.csv \
  -m 10 \
  -l 1 \
  -3pcc 127.0.0.1:4002 \
  10.10.4.131:5080

For each SIPp process, will be used

sipp -i 10.10.4.100 ...
sipp -i 10.10.4.101 ...
sipp -i 10.10.4.102 ...
sipp -i 10.10.4.103 ...

To reach this goal, we use the xargs utility

# Run in parallel
seq 1 "$instances" | xargs -P "$instances" -I{} bash -c 'run_sipp.sh "$@"' _ {}

In this case, all SIPp processes started simultaneously and randomized the CSV file using the same sequence (CSV file contains RANDOM as the first string). As a result, we have a lot of duplicated calls from different IP addresses.
It looks like SIPp processes a randomized string in a CSV file using the current time. If I add dandom delay between SIPp process start, then the generated calls looks random.

Could you randomize the string in the CSP file, not using a timestamp, but data from the /dev/random device? And if this device is not available, then fall back to a timestamp.