[lug] Userland transfers

[mohan kumar]

--- Nate Duehr <nate at natetech.com> wrote:
> On Wed, Apr 18, 2001 at 10:19:13PM -0700, mohan
> kumar wrote:
> > hi guys,
> > I saw a year old discussion in linux-kernel saying
> > something like 100% packet sniffing (and moving
> all
> > the packet data to kernel)is hard to achive in
> linux.I
> > am writing a sniffer application(im no craker...im
> > doing this for my company :-)) which registers
> itself
> > for packet notification, and when a packet arrives
> i
> > copy it to user land for more processing. Will my
> > strategy work with full fledged, duplex, 100MBps
> > networks? Or should i make some hacks in the
> > core/net.c,mmap(for faster userland transactions)
> and
> > do a kernel recompile? which of these strategies
> is
> > good? 
> > (Anyway iam sure what ever strategy i use will
> give
> > good performance than windows, which is in dislike
> to
> > my boss who stands by windows :-( and we end up
> > fighting.Huh)
> 
> You may be reinventing the wheel -- if there's
> kernel level support for
> capturing packets, surely that will always be more
> efficient then a
> user-space application trying to do the same thing.
> 

my app will be a kernel module doing only kernel to
user space transactions :-) My user space module only
does the processing of packet, non-intrusively.

> I think also the 2.4 kernels have the ability to
> pass packets directly
> to user-space anyway -- so either half your code is
> done for you
> already, or you could focus efforts on helping with
> kernel patches to
> fix any speed issues you see there.  (?)

i have to look into 2.4 kernels more deeply. Thanx :-)

Mohan S

__________________________________________________
Do You Yahoo!?
Yahoo! Auctions - buy the things you want at great prices
http://auctions.yahoo.com/