Re: Re: [PATCH] e1000 poll behavior

[Robert Olsson <Robert.Olsson@xxxxxxxxxxx>]

jchapman@xxxxxxxxxxx writes:
 > Is there any advantage in exiting NAPI polled state as soon as
 > possible with the work_done < work_to_do test? Why not stay in
 > polled mode until no rx or tx work is done, i.e. only exit polled
 > state if work_done is zero? 

 Yes it's a variant.

 > When working on e100 a while ago, I found better and more consistent
 > pps numbers by staying in polled mode until the interface went idle.
 > See current e100 code. Unfortunately I don't have e1000 hardware to
 > test if the same is true for e1000.
 
 I'll try this with the forwarding setup when I got a chance but I'll 
 think server-like setups are more interesting to test.

                                                --ro