The MPLS WG Archive

[Shahram Davari <Shahram_Davari@pmc-sierra.com>]

Curtis,

You keep talking about overloading the egress, due to increase in the number of
LSPs that need OAM and due to the misbehaving ingress LSRs.

There are many ways to avoid this:

1) Could implement the OAM processing in HW at wire rate for
all LSPs (same as what you said for ICMP)
2) Could configure (or negotiate) only a limited number of LSPs
to have OAM.
3) Could identify misbehaving ingress, and filter it.
etc.

So processing the OAM packets at the ingress is not the only possible
method. And in fact due to reasons mentioned processing it at ingress has disadvantages,
such as 2x BW usage, variable recovery time, etc.

-Shahram

> -----Original Message-----
> From: Curtis Villamizar [mailto:curtis@workhorse.fictitious.org]
> Sent: Tuesday, December 18, 2001 1:15 AM
> To: Rutemiller, John
> Cc: mpls@UU.NET
> Subject: Re: MPLSOAM BOF meeting draft minutes 
> 
> 
> 
> In message 
> <EB6D4918A175D311971E00204840E2820ABD3B89@whq-msgusr-01.pit.comms.ma
> rconi.com>, "Rutemiller, John" writes:
> > 
> > If it is not behaving properly, why should I expect it to reduce its
> > sending rate?
> 
> If its MPLS-OAM that's a problem.  In ICMP, the egress barely looks at
> the packets shuffles some header bits and sends it back to the sender.
> The errant sender only overloads itself.
> 
> If the router was truly malicius, it would be better served adding and
> withdrawing link adjacencies (as routers are sometimes prone to do).
> 
> > What device is going to report that the source is 
> misbehaving? Only the
> > egress can report this problem.
> 
> You are assuming the sender is completely brain damaged but is able to
> keep all protocol adjacencies going and otherwise go unnoticed.
> 
> > > If an egress is overloaded with OAM, there may be *NO* errant LSP,
> > > just too many perfectly fine LSPs each sending just a 
> little too much
> > > probe traffic.  There are limits to what you can get a 
> microprocessor
> > > to do in a given amount of time.
> > 
> > If the processor is only capable of processing N OAM 
> messages per second,
> > and it enables OAM processing on more LSPs than it can 
> handle, it gets
> > what it asked for. If it only allows as many OAM streams as 
> it can handle,
> > then only errant LSPs will overload it.
> 
> Doesn't it make sense to enable OAM for all revenue bearing LSPs?  If
> so, the ISP does so and the router needs a way to handle this if an
> overload inadvertently occurs.
> 
> > And besides, if ICMP can use a probe as the source, why 
> can't MPLS-OAM
> > also use a sink? The MPLS-OAM label can be switched just 
> like anything
> > else. I can just as easily switch it externally to a sink 
> device. This
> > will provide compatibility with existing equipment.
> 
> A probe refers to a "probe packet" not an adjunct spacecraft attached
> to the ingress LSR that generates ICMP particles.
> 
> The dumb CPU in the ingress generates ICMP echo requests and gets echo
> replies.  Since it determines its own load, non-uniform router
> dumbness does not become an issue.  In other words, if some router
> CPUs are older and slower than most, the slower ones don't fall over.
> 
> > John
> 
> Curtis
>