CCIE-Type Routing Question

[lamav]

Please view the attached drawing before you read on and you may want to leave it open to refer to it.

Anyway, here is the scoop.

The OSPF area represents our Australia network.

Routers SING and JAPAN are OSPF ASBRs that redistribute OSPF-learned routes into BGP.

Both the SING and JAPAN routers form iBGP neighborships with both FMC and ECC in the United States.

So, FMC learns OSPF Australia routes through iBGP from both SING and JAPAN.

And ECC learns OSPF Australia routes through iBGP from both SING and JAPAN, too.

Right now, both FMC and ECC routers favor the SING router as the "next hop."

The reason is that all things being equal, the SING router carries a better IGP metric into its BGP advertisements than the JAPAN router does.

So, if one does a "sh ip bgp 19.210.108.16" (this is an Australia network) in the FMC and ECC routers, they will both show that they have learned 2 routes through iBGP, one from SING and one from JAPAN, but the SING router carries an IGP (OSPF) metric of, say, 330, while the JAPAN router advertises an IGP metric of 445. Therefore, the SING route is favored, so the FMC and ECC routers place the SING-learned iBGP routes in their BGP table.

<i>ecc#sh ip bgp 19.210.108.16

BGP routing table entry for 19.210.108.16/29, version 31730939

Paths: (2 available, best #1, table Default-IP-Routing-Table)

Advertised to update-groups:

2 3

Local

19.213.1.130 (metric 1466112) from 19.213.1.130 (19.213.1.130)

Origin incomplete,

[rpfinneran]

I would simply set the local preference at the Japan and Sing routers during the redistribution, since they are in the same AS as the fmc and ecc routers. You could do this using a route-map to ensure it only affects traffic destined for Australia, and no other connections that my not be shown in your drawing. Good luck

[marikakis]

As I have said before, local preference causes a lot of traffic to move in one direction only. The SIG link to Australia has utilization issues. If someone sets the local preference to move a lot of traffic towards the JAPAN link to Australia, in the best case they are effectively moving to the near future the utilization issue from the one link to the other. There are utilization issues in this network and the question is not just how to route the traffic, but also to consider all the link utilizations. This is the vital information that has not been provided by the client from the beginning.

If they are now choosing closest exit routing by restting the MEDs, they are choosing a more long term solution that can be fine tuned. What would they do if the other link starts experiencing the same issues in the future (if not already experiencing or some other link does)?

Also, some large networks have well defined policies and handle higher level routing decisions with local preference (such as choice between client, peer, upstream routes), and would not be willing to have different local preferences here and there when there exist other ways to do load balancing.

I guess local preference is elegant when you have a couple of routers or link utilizations are not considered, as is usually the case with exams. Although this looks like an exam question from the title, it actually refers to a real network with link utilization being an issue. Another cool thing with exams is that no angry customers are calling while you are trying to set up the network and you have hours to save the world, while in a real network you might have only a couple of minutes or seconds. In those minutes, it is easier to think about the bandwidths and the utilizations of links, than trying to remember what local preference you have set for each route, where in the network have you set it and why (an approach clearly unmanagable).

[marikakis]

Guiseppe,

I would really like to read your opinion on this last point, if you have the time, since you deal with big customers and you typically have something good to say in such issues. I feel like a lone rider in here, although I am thinking that if I was really wrong, you would have stepped in.

Kind Regards,

Maria

[Jon Marshall]

Maria

No Giuseppe but hope you don't mind a few comments. I have always found your posts to be very helpful and full of technical detail. And i also consider you to have more knowledge than i do with BGP. In addition as far as i am aware nobody is saying you are wrong, as often with these sort of things right and wrong don't come into it.

But the original problem was how to send all traffic via JAPAN, that was the customer requirement. It was not how to load-balance between the 2 links. If it was then i don't think any of us would have suggested local preference as the simplest solution.

With the people involved in this thread i have never seen them assume that people do not know what they are talking about and i would hope i haven't either. So to say we are underestimating the engineers knowledge is wrong. Perhaps this engineer has no or limited experience with BGP and is a lot more comfortable with EIGRP. Perhaps that's why Victor is being asked to look at the issue as well.

Then again, perhaps the engineer sees something we don't, perhaps Victor has been given incomplete requirements, perhaps as you say the engineer has taken for granted a lot of things about the network that he has not passed on. But they are all perhaps.

When you say "I am not the type of person that insists on something just so my own solution is the selected one (and so I get the points in this case)!"

that implies that there are people who do this. Again i would say as far as this post is concerned i am not aware of that happening although perhaps you feel differently. I agree 100% with you as i'm absolutely sure the others do ie. the important thing is to get the right answer.

We all use our experience as well as knowledge to try to help people but we all have different experiences.

Jon

[marikakis]

Jon,

I know it might look like I am assuming a lot in this case, but sometimes you can call this instict. Instict is typically associated with women, but is essentially a quick analysis of various factors that you cannot phrase at the time your brain processes all the available information. I had a feeling there was some load balancing requirement in this, as is typical in cases of 2 links with one overflowing. Also, Victor said at the very beginning that he is not familiar with the network yet or what the client is trying to, asked what people think although he already got answers, and then said I was wrong. I still cannot understand this. How can someone be sure I was wrong at this point? Unless the idea was me to confirm that he was right (although he doubted that himself) and I didn't catch it.

I won't insist on anything, although when many solutions exist, one can be more managable and scalable than the other. This is not my network after all and if anyone insists on anything they should be ready to accept the consequences of what they are doing. Sometimes people in the forum say things in an academic fashion (I know I have done that mistake as well), only because they feel more comfortable when their own network is not discussed. Still, I sometimes feel like the network is really mine and try to think of what could go wrong. And yes, people can insist just to get the points. I say what I think and I don't care about that, so I don't like the other person to think that of me and ignore me when I am only trying to help.

Thank you for your reply. I don't feel a lone rider anymore and I appreciate you answering because Guiseppe might not catch the thread after becoming so long. I do appreciate your answer. I asked Guiseppe because we typically meet in such cases and he has no problem telling me I am wrong in something :-)

Kind Regards,

Maria

p.s. To add to the instict: it is typical for people to be confused with the difference between metrics and MED even if they know BGP well enough, so I thought this guy was only missing just one detail to do is job.

p.s.2 I just rated your post, although I usually don't say this when I do it for various reasons. I just noticed that you thought I was referring to someone in this thread about the points, which is clearly a misunderstanding. Jon, I have to say you are underestimating me now :-) I know what you guys are doing in this forum!

[Giuseppe Larosa]

Hello Maria, Jon,

I have been out all the afternoon.

As Jon has written no one here thinks to be better then others, or assumes no knowledge of the subject on the other people.

I can say that Maria's posts are usually high quality posts with in depth analysis.

These posts make me think I should write less posts of better quality.

I think Victor had few details about network and customer needs, and he was under pressure to provide a feedback.

He was puzzled for the customer engineer proposal of changing EIGRP metric because he feels it cannot influence current best BGP path choice, at least without other changes in BGP attributes of involved routes.

And everyone here agrees on this.

I suggested to use local-preference, Victor asked for other opinions and he has received Jon's answer and your posts.

I think that Victor has just tried to say that he had no time to follow your suggestions (as he had no tim:e to discuss how OSPF metrics are carried inside BGP :) ... )

Be aware that this doesn't mean you (Maria) are wrong in your "instinct": what you suggest can be the real customer needs/objectives or it can be the next customer needs/objectives if the Japan link becomes overwhelmed.

As a last note:

local-preference can be modified on a per prefix basis so some prefix based administrative load sharing can be implemented in order to achieve the desired level of usage of the two exit points.

if the BGP MED are made equal and the IGP metrics in the EIGRP domain are used to decide how to use the exit points there is no added value in having implemented an iBGP full mesh: it is like having redistributed OSPF routes directly in the EIGRP domain.

Again Jon had noted in one of his first posts that we don't know why the customer has implemented BGP.

Once that iBGP has been deployed and it is in use it is common practice to use its rich set of tools to influence routing paths.

Hope to help

Giuseppe

[marikakis]

I can only say I really love you guys! :-)

p.s. I have to work on my thesis at this point, so I try not to start conversations I am not ready to complete. That's why I post less :-)

[lamav]

Folks:

This is no place to be thin-skinned. There are a lot of ideas, comments and suggestions going back and forth and a lot of good things being "said" by a lot of people. Therefore, its not always eays to entertain everyones suggestion or comment extensively on everyones opinion. Someone asks for opinions, you give it, and they either acknowledge it or not; that's it. No need for long, drawn out explanations about men, women, etc etc etc....its all inconsequential.

As for the network, yes, my knowledge of this particular segment of the network --- especially at the time I introduced this topic -- wasnt too great. I had never seen any of these routers or this part of the clients network before. I made that perfectly clear in my initial post, or shortlay afterward.

Moreover, the client did not specifically ask for MY feedback. But, I learned that the client wanted to achieve a certain affect and that he was going to proceed in what seemed like a sub-optimal manner, so as a consulatnt I took it upon myself to offer some advice. I also made it perfectly clear that I had not had a chance to ask the client exactly what he was thinking and why he wants to proceed that way. I did have my suspicion, though, that the plan was not even hatched by an engineer, but by a project manager who was being adventurous and perhaps oversimplifying things. So, I proceeded pro-actively to see what the plan was and I also gave my opinions on it. No biggie.

I posted my scenario on this board because I know there are some pretty damn smart people on here with a lot of experience: Jon, Giuseppe, Rick Burts, Edison, Harold Ritter, Pablo, Joseph Doherty, and many more, so I wanted to hear a variety of opinions, including Maria's, who until this thread, I actually had never seen before. But Im glad she gave her opinion.

[marikakis]

Victor,

I do consider my pure technical posts better than the others, so I knew many of my posts here were destined to not be liked even by me :-)

I guess I caught some fire when you called me a young lady! You probably meant well, but as we have seen before in this case, non-technical comments can cause misunderstandings.

Thank you for opening this case, as I always have fun with such cases no matter what.

Kind Regards,

Maria

[lamav]

OK, here goes.

I just got off the phone with the client and here is the scoop.

What they want is to deliberately install some asymmetric routing into the mix.

Traffic from Australia and destined for ECC and FMC (U.S.) will always take the route through SING because it is advertising a default route and there are no specific routes in the Australia routing tables for the US. SING and JAPAN are ASBRs and they both inject default routes into Australia, but the better metric is through SING.

On the other hand, they want to manipulate the delay to force traffic destined for Australia from the US to go through JAPAN, as we discussed all along (now we're talking about the other direction).

The changed the delay on the ECC and JAPAN routers, and now some of them Australia-bound traffic goes through JAPAN, but most of it still goes through SING. So what they want to do is further this solution by maniipulating the seed metric in OSPF in Australia BEFORE its redistributed into BGP so that BGP can pass the metric they want to the US through its BGP neighborships.

Personally, I think this is a convoluted approach. I would not deliberately install asymmetric routing and bother manipulating the metrics in such a seemingly random fashion to make it work. I would manipulate BGP atttributes, as we discussed before.

What do you guys think of their solution?

[Giuseppe Larosa]

Hello Victor,

I agree I would use BGP attributes they are acting as BGP wasn't present.

Unless they are also planning to remove BGP this approach requires more work and provides less control.

With BGP you can easily decide on a per prefix basis what would be the preferred exit point.

Playing with IGP metrics has a greater impact: when it works it can move all traffic to that exit point.

I'm not surprised that most traffic is still going via Sing router ...

Hope to help

Giuseppe

[Mohamed Sobair]

Hi Vector , ALL,

There are multiple approaches to achieve thier desired goal.

I would personally choose to modify the Local preference in Japan router to influence outgoing path On the IBGP dommain.

Still modifying Eigrp delay at SING router facing FMC, I can see that they are 2 links used by Eigrp , Modifying the delay would also reflect the total metric recieved by ECC and Japan, but would require additional config if you want to achieve loadbalancing the traffic across both serial interfaces, In other word you will have to use the "variance" command, Increasing the Delay at SING router is another approach, ECC and Japan router would then recieve the path to network 19.210.x.x with higher metric than japan does, thus prefering japan Route, but the easiest way would be modifying the local preference at Japan router.

As Maria noted earlier, what if they want in future to chang the path or if the path gets overwhelmed? IS there any sort of loadsharing here? What are the exact Network redistributed by ACR03 and ACR01? How (SING and SING2), (SING2 and Japan) connected?

HTH

Mohamed

[lamav]

Mohammed:

Long time...

Well, they are viewing this as a temporary solution.

They want traffic bound for Australia to go through JAPAN and traffic in the other direction to go through SING.

So they reduce the delay on ECC and JAPAN to force ECC and FMC to go through JAPAN to get to Australia.

So their idea of load balancing/sharing is to have traffic use one route in one direction, and a separate route in the other.