Re: [OPENETLAB] Question: Oplets and Accelar Queue Mgt./Routing

[Rob Jaeger <rfj@cs.umd.edu>]

On Thu, 21 Sep 2000, Phil Wang wrote:

> 
> 
> "Sawant, Vivek" wrote:
> > 
> > Rob Duncan wrote:
> > >
> > > "Sawant, Vivek " <vivek@americasm10.nt.com> writes:
> > >
> > > > I am playing with an idea of influencing Accelar Forwarding from Oplets.
> > > > I am looking at an algorithm which is an extension to RED. I wonder if I
> > > > can implement it as an Oplet and combine it with Accelar IPfilter
> > > > setting to influence Accelar Packet forwarding.
> > > >
> > > > My initial investigation suggests that this is not possible. Here is
> > > > why. This algorithm, like any other queue management algorithm, includes
> > > > the just arrived packet in a computation to decide if the packet should
> > > > be dropped or queued. If this computation happens in an Oplet, it is
> > > > essentially happening outside the Accelar forwarding path. Given current
> > > > Accelar filter actions, one can either choose to drop or forward a
> > > > packet (besides getting a copy of it, etc.) One cannot pull a packet
> > > > out, do computations and then give it back to Accelar forwarding engine.
> > > > (I know, it will radically slow the engine down.) So, such a queue
> > > > management algo has to be part of the Accelar forwarding mechanism.
> > > >
> > > > Is my inference correct or is it still possible to implement such a
> > > > scheme through Oplets? Does Accelar support RED queues?
> > >
> > > Yes, I believe you are right.  We don't currently have an inverse
> > > operation for JCapture.  Right now the only way to send packets out of
> > > the ORE is by using a socket, which of course is not what you want.  I
> > > think OpenIP does have the kind of API you want, so we shoudl take a
> > > look at it.  I don't think it's any harder to implement that JCapture
> > > is...
> > >
> > > R.
> > 
> > Let me understand this. So, there is a way to intercept an incoming
> > packet
> > and send it back to Accelar forwarding engine? If this is true, it's a
> > great news. I had almost lost hope for my particular experiment.
> The answer is yes at a price of memory and performance. See my last
> email.
> 
> > 
> > Where can I look at the OpenIP API mentioned by you. Is there anything
> > I can do to help implement support for this in ORE?
> http://openip.ca.nortel.com.

this url is stale.