netgraph locking / performance [was: ... AOE]

[Harti Brandt]

On Mon, 6 Sep 2004, Bjoern A. Zeeb wrote:

BAZ>On Tue, 31 Aug 2004, Scott Long wrote:
BAZ>
BAZ>Hi,
BAZ>
BAZ>> My employer has done extensive profiling of packet delivery through
BAZ>> netgraph.  While the locking of the netgraph framework is definitely
BAZ>> correct, it's not terribly efficient and leads to a good deal of
BAZ>> latency.  We are looking at various proposals on how to address this.
BAZ>> This isn't a criticism of you or Netgraph, just a set 'real-life'
BAZ>> observations under very high load (bridging and packet inspection on
BAZ>> 4 GigE links simultaneously qualifies as high load =-)
BAZ>
BAZ>could please explain a bit more / give some numbers ? Or are there
BAZ>any published results ? What do you mean by 'packet inspection' ?

That would also interest me. I did measurements with my satellite link 
simulator (a netgraph node) and got a couple of 100usec latency when 
pushing 150k ATM cells through the node (each cell counting as a packet). 
including the processing of the two ATM adapters. Does this already count 
as high latency?

harti