GitLab

A customer is reporting a depressingly low performance with proxy MySQL. The numbers they get are radically different compared to what we get in perflab.

The usage scenario is a CMTS rebooting 4000 cable modems (clients) behind it asking for leases. The goal of this mode is to measure how long it takes to provision specified number of clients. An important factor here is that the clients that don't get a response within a second start a retransmission (search for exponential backoff in RFC2131).

I can imagine something like this:

./perfdhcp --provision-clients 4000
...
...
It took 123.456s to provision 4000 clients. 5678 packets were sent, 1234 retransmissions needed, received 8000 responses.

This is only a proposal. The feature may look and work differently. The goal is to have something better than we have now to simulate actual real life scenario of mass client reboot.