Oliver Yang writes: > Andrew Gallatin wrote: > > > > Excellent response. If you want to test an example of this, run any > > sort of client/server TCP streaming benchmark (netperf, iperf, etc) > > where the receive side is coming across your driver, and use a large > > TCP window size. Then suspend the server side of the benchmark > > while traffic is still flowing. Then try to remove your driver. > > The system should not panic :) > > > Yes, it's another our regular test case for NIC drivers. :-)
That reminds me that I forgot the corallary: Make sure your driver still passes traffic after you've got enough non-consuming applications hogging your desballoc-loaned private buffers. Ie, make sure that your driver does not loan out all its buffers, but copies into allocb'ed buffers when it starts to run short of desballoc-loaned buffers. Drew _______________________________________________ networking-discuss mailing list [EMAIL PROTECTED]
