Re: [nsp] Cisco vs. Juniper of LSP setup

From: Eric Osborne (eosborne@cisco.com)
Date: Fri Apr 12 2002 - 16:30:07 EDT


On Fri, Apr 12, 2002 at 01:10:34PM -0700, LU wrote:
> I may be missing some thing here.
>

Yes...
 
> If there are multiple IGP paths and only one of them
> has the bandwidth demanded from the LSP, even if
> without TE-database, we should be able to get there,
> just try them one by one till it finds it.
>

How do you propose to find the right path? What decision algorithm
will you use that will find the best path for a given LSP without
being able to see the path in front of it?

The problem with that is that you either need crankback or explorer
frames, neither of which are particularly pleasant (or standardized
for TE). Plus, since you'll tend to have far more TE LSPs than
nodes/links, you'll end up with more traffic exploring that you would
just distributing info via IGP.

If you want to get from your house to a place you only know the
address of, but not directions to, do you just drive down the street
and keep turning left until you find the place? that'll work, but
it's not terribly efficient. That's why we have maps....
 
> If I want to a static LSP, I can just define the path
> and bandwidth, if it can be accepted or not is another
> story which depends on my estimation of the path
> status. Is that how people calculate LSPs off-line in
> stead of on-line?

No....offline calculation defines explicit paths for LSPs to take.
You're talking about not propagating IGP information and letting the
network magically find the right path. Bad idea; not only does it not
scale, but even in small, infallable networks, you can still get into
race conditions that cause (at the very least) more signalling churn.

>
> I think wether static is useful or not, it should be
> up to the users, I guess that's why we still have
> static and dynamic routing protocols.
>

With IOS, you can declare an explicit path, which is a path where
every hop is defined. The explicit path is checked against the TE DB
(excepting inter-area TE, which is a whole different conversation),
and the signalling is sent out only if there's a reasonable chance of
making it. Explicit can specify links and/or nodes, and unless you
specify every link an LSP takes, you still need the IGP DB to figure
things out. JUNOS does something very similar, if not exactly
similar; explicit paths are a necessary piece of any TE
implementation.

Explicit makes more sense than no-cspf/verbatim. Flooding information
in the IGP is necessary to deal with the reality that things are
dynamic. Static unchecked paths are useful in corner cases, lab
trials, and networks where you have more signalling than you'd have
TE; in the case where you'd have more signalling than TE, I submit you
don't need MPLS-TE at all.

eric

> Thanks
> LU
>
>
>
>
> --- Eric Osborne <eosborne@cisco.com> wrote:
> > On Fri, Apr 12, 2002 at 11:37:19AM -0700, LU wrote:
> > > Eric,
> > >
> > > I am just comparing how JUNOS and IOS are
> > different
> > > regarding this. My concern is that enabling OSPF
> > > opaque will generate too many additional LSAs, I
> > am
> > > not sure if that will be a problem, but that's my
> > > concern. If I only want to create a LSP tunnel,
> > why do
> > > I have to use opaque LSA? Why can I just use
> > non-TE
> > > database to set up the tunnel?
> >
> > This is like asking why you have to use an IGP, and
> > can't just use
> > static routes everywhere. The only way to have
> > anything useful is to
> > do it dynamically. What happens if a link in your
> > verbatim path goes
> > down?
> >
> > > Sure, the TE-database will make the setup faster,
> > but I still do not
> > > see why trying to reserver bandwidth on a no-cspf
> > path is a problem.
> >
> > consider:
> > F
> > / \r
> > D \r
> > / \r
> > / G |
> > A---B---C > J
> > H |
> > / /
> > E /
> > /
> > I
> >
> >
> > - there is a verbatim LSP from A to J that wants 3
> > units of BW
> >
> > A-B and B-C have 10 units available
> > C-D has 5
> > C-E has 10
> > D-F has 0
> > D-G has 0
> > F-J has 10
> > G-J has 10
> > E-H has 3
> > E-I has 4
> > H-J has 10
> > I-J has 0
> >
> > all links have equal IGP cost.
> > there are therefore 4 possible paths your signalling
> > can take
> > Only one (A-B-C-E-H-J) can fit this LSP.
> > How do you ensure the LSP takes the right path
> > without being aware of
> > the link bandwidth in the network? Do you configure
> > 4 verbatim paths?
> > What happens if you have an even more complex
> > network, with more than
> > 8 nodes? What happens if you have multiple TE
> > tunnels in this
> > network?
> >
> >
> > > Bottom line is, I just want to create a
> > > static or dynamic LSP, I do not care about any
> > other
> > > information, thus the opaque LSAs are useless for
> > this
> > > task, and they will create more instability.
> > >
> >
> > This is great in a network that's either very small
> > or which will
> > never have link/router flaps. So are static routes.
> > If you have a
> > network where verbatim will really work for you,
> > even under various
> > failure conditions, feel free to use it. While it's
> > true that we
> > don't have a knob as simple as no-cspf, I have yet
> > to see a pressing
> > need for it in real life. I'm sure you can find
> > knobs that either
> > vendor has which the other does not, but which are
> > pretty useless.
> >
> > For example, we have a knob, 'mpls traffic-eng
> > link-management timers
> > bandwidth-hold' that I've been told JUNOS does not.
> > I have yet to
> > find a use for this knob.
> >
> > It sounds to me like you're taking an academic
> > approach to this,
> > rather than a real-world one. You're free to do
> > that, but if you do,
> > you will run into all sorts of 'shortcomings' that
> > are academically
> > useful but operationally nonsensical.
> >
> >
> >
> > eric
> >
> >
> >
> > > Thanks
> > > LU
> > >
> > > --- Eric Osborne <eosborne@cisco.com> wrote:
> > > > On Fri, Apr 12, 2002 at 10:49:49AM -0700, LU
> > wrote:
> > > > > Sean,
> > > > >
> > > > > If I understand this right, not only IOS does
> > not
> > > > have
> > > > > the function of ignoring TE datatbase like the
> > > > > "no-cspf" in Juniper, it also can not run RSVP
> > > > without
> > > > > enabling OSPF opaque LSA on a interface. I
> > have
> > > > not
> > > > > tried IS-IS, but I assume it has to have the
> > > > > wide-metric enabled in order to run the
> > MPLS-TE,
> > > > > right?
> > > >
> > > > There is no command to explicitly enable/disable
> > > > opaque LSAs in TE.
> > > > There are two commands you could be referring to
> > -
> > > > 'mpls traffic-eng'
> > > > under OSPF, or 'mpls traffic-eng' on the
> > interface.
> > > > Both do more than
> > > > just opaque LSAs, tho.
> > > >
> > > > Why do you want to do this? It works fine in
> > some
> > > > cases, but if you
> > > > have a no-cspf path and you try to reserve
> > bandwidth
> > > > on that LSP, you
> > > > can paint yourself into a corner.
> > > >
> > > > There are some legitimate uses of
> > verbatim/no-cspf,
> > > > but I'm curious to
> > > > see what your use is.
> > > >
> > > >
> > > >
> > > > eric
> > > >
> > >
> > >
> > > __________________________________________________
> > > Do You Yahoo!?
> > > Yahoo! Tax Center - online filing with TurboTax
> > > http://taxes.yahoo.com/
> >
>
>
> __________________________________________________
> Do You Yahoo!?
> Yahoo! Tax Center - online filing with TurboTax
> http://taxes.yahoo.com/



This archive was generated by hypermail 2b29 : Sun Aug 04 2002 - 04:13:40 EDT