> ifspeed is the kstat that has been used for decades now, although not 
 > always consistently.  (Sometimes the units have been Mbps.  Sometimes 
 > they have been just bps.)
 > 
 > Again, we need to just pick something, and stick to it.

I agree.  However, given how broken and inconsistent we've been in the
past, I think we should prioritize clarity over all else.  To that end, I
would avoid "ifspeed" since the "if" prefix is often used (for Solaris
anyway) to refer to things at the IP layer, and avoid "link_speed" because
the "link_" part is redundant and annoying to type.  In fact, it was this
thinking that made me arrive at "speed" when doing the WiFi work.

 > Frankly, at this point, I'd really like to see an implementation advice 
 > to developers about what stats to report, and they should represent.  
 > (For example, what counts as an ierror?  What counts as a macrcv 
 > error?   What is the difference between first_collisions, 
 > multi_collisions, ex_collisions, and collisions?  What about collisions 
 > detected on rx instead of just tx?  Etc.  Should opackets/obytes count 
 > packet transmission attempts, or only successfully transmitted packets?  
 > Etc.

Sounds like a nice manpage in the making.

-- 
meem

Reply via email to