The MPLS WG Archive

Cell Relay Retreat>MPLS WG Archive>month:2000-Jun> msg00478



[Date Prev][Date Next][Thread Prev][Thread Next]  
  [Date Index][Thread Index][Author Index][Subject Index]

TE

  • From: "alex.mondrus" <alex.mondrus@ipoptical.com>
  • Date: Fri, 23 Jun 2000 12:26:30 -0400
  • X-Sent: 23 Jun 2000 16:18:17 GMT

Mike:

Please review the following document

http://www.seas.upenn.edu/~guerin/publications/qos_ospf_rfc.txt

Alex Mondrus

http://www.ipoptical.com




-----Original Message-----
From: owner-mpls@UU.NET [mailto:owner-mpls@UU.NET]On Behalf Of Shahram
Davari
Sent: Friday, June 23, 2000 11:31 AM
To: 'Mike Badil'; mpls@UU.NET
Subject: RE: TE


Hi Mike,

In theory it is possible to do some sort of limited constraint-based routing
without Explicit-Routing (with or without MPLS). In order to do that you
need to define a new routing protocol (let's call it QoS-OSPF), which can be
run on all routers and which can compute the best path according to some
predefined constraint. Drawbacks are:

1) All nodes need to do CR computation.
2) You can only support one set of CR rules.

Cheers,
-Shahram

>-----Original Message-----
>From: Mike Badil [mailto:hasko10@hotmail.com]
>Sent: Wednesday, June 21, 2000 8:49 PM
>To: mpls@UU.NET
>Subject: Re: TE
>
>
>
>Hi
>
>>True - you can do that in IP too.
>>
>>However, in IP if you want to do explicit routing, you need
>to include the
>>full
>>path information in each IP packet header. Now, that will
>dramatically
>>increase
>>the IP packet header size and consequently, the processing time will
>>increase.
>>The explicit path used in MPLS has much less overhead
>compared to that in
>>IP.
>>"alex.mondrus" wrote:
>
>Yes, you are right, explicit routing is clear. I agree with what you
>said,but My question was about Constraint Based Routing. How to choose
>explicit path(best path according to TE)?, that explicit path is not
>shortest path which calculated by OSPF.
>
>
>
>
>Thanks
>
>
>>
>> > In general to do an explicit path set up you do not need
>to use MPLS.
>> >
>> > Alex Mondrus
>> >
>> > http://www.ipoptical.com
>> >
>> > Traffic Engineering - MPLS allows you to set up tunnels
>> > >         with explicit routing so that you can design your routes
>> > >         off-line in order to achieve best traffic distribution to
>> > >         minimize network hot spots (which often occur in OSPF
>> > >         type routing) and improve network capacity efficiency.
>> > >         Load balancing (parallel MPLS tunnels running between
>> > >         a source-dest pair) and trunk protection also
>become easy.
>> >
>> > -----Original Message-----
>> > From: owner-mpls@UU.NET [mailto:owner-mpls@UU.NET]On
>Behalf Of David
>> > Wilder
>> > Sent: Wednesday, June 21, 2000 12:02 PM
>> > To: Jay Wang
>> > Cc: hasko10@hotmail.com; mpls@UU.NET
>> > Subject: Re: TE
>> >
>> > Very well put Jay.
>> >
>> > - Dave
>> >
>> > > Mike,
>> > >
>> > >     MPLS is not created to support constraint based routing
>> > >     and it doesn't.  It let you set up layer two tunnels using
>> > >     label switching.  An IP network can benefit from the
>> > >     existence of MPLS with at least the following:
>> > >
>> > >     1. Faster data forwarding at the transient (core) routers -
>> > >         This is because there is only label (index) lookup and
>> > >         no IP header processing.  This however became much
>> > >         weaker an argument lately since lots of work that used
>> > >         to be done by software (e.g., header processing,
>> > >         classification) now are typically done in ASIC
>in vendor's
>> > >         box.
>> > >
>> > >     2. Traffic Engineering - MPLS allows you to set up tunnels
>> > >         with explicit routing so that you can design your routes
>> > >         off-line in order to achieve best traffic distribution to
>> > >         minimize network hot spots (which often occur in OSPF
>> > >         type routing) and improve network capacity efficiency.
>> > >         Load balancing (parallel MPLS tunnels running between
>> > >         a source-dest pair) and trunk protection also
>become easy.
>> > >
>> > >     3. QoS/VPN  - MPLS interworking with Diffserv gives
>you traffic
>> > >         isolation (and hence some performance protection). Also,
>> > >         MPLS with proper support of resource
>> > >         reservation signaling mechanism (e.g., RSVP),
>you can specify
>> > >         thE size of each MPLS 'pipe'.  With a careful
>traffic trunk
>> > > analysis,
>> > >         you may set up a set of corresponding pipes
>(using constraint
>> > based
>> > >         or explicit routing) such that you can place
>some application
>> > > specific
>> > >         (e.g., voice,  video) flows on the trunks while
>meeting some
>> > >         stringent real-time specs (e.g., loss, jitters, latency).
>> > >
>> > > - Jay
>> > >
>> > > David Charlap wrote:
>> > >
>> > > > Mike Badil wrote:
>> > > > >
>> > > > > I just reading MPLS document, I have a question
>which I am not
>>really
>> > > > > clear to understand, if someone helps I will be happy.
>> > > > >
>> > > > > The Question is:
>> > > > >
>> > > > > What makes MPLS to support constraint based routing?
>> > > >
>> > > > The existance of LSPs makes traffic engineering easier
>to implement.
>>  It
>> > > > does not make it impossible, however.
>> > > >
>> > > > > If it just adding contsraint metrics to conventional
>routing in
>>order
>> > > > > to support TE.
>> > > > > Why it is not possible without MPLS.
>> > > >
>> > > > Sure, it's possible.  Who said it wasn't?
>> > > >
>> > > > > In other word, in conventional IP routing, if we add
>constraint
>> > > > > metrics, can conventional IP routing support it also?
>> > > >
>> > > > I think you're missing the point of MPLS.
>> > > >
>> > > > MPLS's purpose is not to create the ability to perform
>>constraint-based
>> > > > routing and traffic engineering where it was
>previously impossible.
>> > > >
>> > > > MPLS's purpose is to create a connection-oriented link
>layer (COLL).
>> > > > Where forwarding decisions are made solely on the basis of a
>>packet's
>> > > > label, and not on any other content in the packet.
>> > > >
>> > > > The use of a COLL is not a requirement for traffic
>engineering.  It
>> > > > simply makes it easier to implement.  With a COLL, the
>hard work of
>> > > > determining the path that data packets must take can
>be done once,
>>when
>> > > > the LSP is set up.  Without a COLL, this work must be
>done by every
>> > > > switch, for every data packet.
>> > > >
>> > > > Connection oriented link layers are not new.  ATM and
>Frame Relay
>>also
>> > > > use them.  The big thing that makes MPLS special is
>that it can run
>>over
>> > > > nearly any transport medium (ATM, FR, POS, Ethernet,
>etc) instead of
>> > > > being tied to a specific layer-2 encapsulation.  Also,
>because it
>>uses
>> > > > IP for its addressing, it can work with many common routing and
>> > > > signalling protocols (like OSPF, IS-IS, and RSVP).
>> > > >
>> > > > -- David
>> > >
>> > > --
>> > > Jay Wang - http://math.research.bell-labs.com/~wang/wang.htm
>> > > Bell Laboratories, Lucent Technologies     Tel: (908) 582-7223
>> > > 600 Mountain Avenue, Room 2C-308,
>wang@research.bell-labs.com
>> > > Murray Hill, NJ 07974-2070
>> > >
>> > >
>> > >
>>
>>--
>>Regards,
>>Arup Das
>>(763-1863)
>>
>>
>>
>
>_______________________________________________________________
>_________
>Get Your Private, Free E-mail from MSN Hotmail at
>http://www.hotmail.com
>



  • References:
    • TE
      • From: Shahram Davari <Shahram_Davari@pmc-sierra.com>