The MPLS-OPS Archive

[Jean Philippe Vasseur <jvasseur@cisco.com>]

Hi,

At 15:52 13/03/2003 +0100, Sebastien.Spas@alcatel.be wrote:
> Hi,
>  
> Off-line TE approach is always
> centralized and requires utilization of a specific tool (a software
> product). The main advantages are :
> -averaging link utilization.
> This is reducing impact on your service when one link goes down, avoid
> congestion, and allows to route 35% more traffic than online TE approach
> (that 35% is actual value of tests on a 16 routers, 80 links and 40
> bi-directional LSPs network we made with 5620 TSOM, the Alcatel MPLS-TE
> product). The final reach is reducing equipment cost by 35% for the same
> ammount of traffic.
> -implementing LSPs protection
> in a centralized way, which is much more efficient
> -the possibility to use lot of
> very interesting features in the offline tools, like performance
> management, support for planning, auto-discovery and provisioning, which
> makes the off-line TE process much more easier, and then can allow to
> start it each week of 2 weeks. This makes the off-line tool much more
> reactive, and allow to perform network planning every week or 2 weeks,
> allowing to follow closer the network traffic evolution, and so allowing
> to reduce the risk marging implemented by over-provisioning. You can them
> in a unique tool perform the traffic monitoring, mandatory for planning
> the LSPs, test different planning scenarios and finalyl implement them,
> with only few clicks in a single product.
> -this approach can also cope
> with multi-vendors networks since it's software based.
>  
> On-line : CSPF online approach
> is directly implemented by the routers, so it's totally transparent, you
> don't have much control about the paths used by the LSPs, and the total
> traffic increase with this approach is pretty poor compares to off-line
> solutions. The on-line TE is not needed as fallback in case of link
> failures. Off-line TE approach allow to also provision backup disjoint
> paths, and supports fast reroute.
>  
> On-line TE approach can be more
> interesting to first provision new LSPs quickly, which are optimized on a
> regular basis with an Off-line tool.
>  

Don't get me wrong ... I'm not trying to argue in favor of one or another
approach, both are perfectly valid and
have their pros and cons. By the way, one could easily negate
several of your arguments but again the point was more to provide an
overview of the applicability of both methods, your view is really
subjective. I would just mention that the number of 35% cannot be taken
for granted and such results are highly topology and traffic matrix
dependent; furthermore your correlation between bandwidth and equipment
saving is also quite arbitrary.

JP.

 
you can find more information about alcatel product there : http://www.alcatel.com/products/productsummary.jhtml?_DARGS=/common/opg/products/include/productbrief.jhtml_A&_DAV=/x/opgproduct/a5620tsom.jhtml
 
 
kind regards,
sebastien.

-----Original Message-----

From: Jean Philippe Vasseur [mailto:jvasseur@cisco.com]

Sent: jeudi 13 mars 2003 15:17

To: Lavoisier José Leite Farias

Cc: Sandmaier, Jan; ILazar@burtongroup.com; mpls-ops@mplsrc.com

Subject: Re: RES: [MPLS-OPS]: On-line TE?

Hi,

At 13:59 12/03/2003 -0300, Lavoisier José Leite Farias wrote:

Is there any advantage by choosing one or other option ? I understood by reading Irwin's answer that on-line TE is used as fault mechanism.

No, on line TE is not used as a fault mechanism, you have the two options:

        - off-line (centralized): provides global optimization as the set of TE LSPs are simultaneously computed by a single entity that makes  use of search mechanisms, to find a placement that for instance tries to minimize the average load utilization of each link (this is just one   example),

        - on-line (distributed): each LSR independently computes paths for its own set of TE LSPs using CSPF. Inability to find placement due to        bandwidth fragmentation can be highly reduced by the use of appropriate mechanisms like (soft) preemption.

Both are currently deployed by Service Providers.

JP.

I believe off-line calculation it seems to keep the whole network optimized. am I correct ?

 

Regards,

Lavoisier.

-----Mensagem original-----

De: Sandmaier, Jan [mailto:Jan.Sandmaier@t-systems.com]

Enviada em: quarta-feira, 12 de março de 2003 11:44

Para: ILazar@burtongroup.com; mpls-ops@mplsrc.com

Assunto: AW: [MPLS-OPS]: On-line TE?

Irwin,

the two concepts are not so different. In off-line TE you collect traffic rates from all your LSPs in the network, feed the traffic matrix in an optimization engine which produces explicite routes which are used by a provisioning tool to build up the new paths. It is up to the oparator when he adjusts the model or which triggers are used for ajustment. Engineered paths are also adjusted accordingly based on actual network conditions as with on-line TE. You may use on-line TE/dynamic LSPs as a fallback mechanism in case of a link outage. The triggers in on-line TE are configured thresholds and reoptimization timers so there is no big difference. The difference is more centralized computation versus distributed.

Jan

-----Ursprüngliche Nachricht-----

Von: Irwin Lazar [mailto:ILazar@burtongroup.com]

Gesendet: Freitag, 7. März 2003 16:32

An: mpls-ops@mplsrc.com

Betreff: [MPLS-OPS]: On-line TE?

I've seen quite a few vendorst that offer sophisticated systems for off-line TE.  My understanding is that this is done by downloading sample network data into a TE optimization engine?  Correct?

Is there anyone out there doing on-line TE - in which real-time network traffic is processed and engineered paths are adjusted accordingly based on actual network conditions?

Thanks,

Irwin