[OpenAFS-devel] .35 sec rx delay bug?

[Jim Rees]

chas williams - CONTRACTOR wrote:

  no idea.  jim had the problem not me.  i run this code locally and i
  havent had any complaints.

I never tested it, so I told Jeffrey to ask you if you had tested it.
Sounds like the answer is yes.

  of course.  the solution is not ideal.  as i mentioned, you would need
  path mtu discovery to implement a completely correct solution.  i would
  personally vote that you should not send datagrams > 1500 off your local
  subnet.  while jumbograms increase performance, its not clear to me 
  that the benefit outweighs the cost when going outside the local lan.

It does if you have large mtu all the way through.  Eventually I think we
really need path mtu.