EIGRP

[rachelhor]

Hi,

Have a few queries hope someone can clarify and advice.

Q1) Pls see attached file "Configuring EIGRP" page 1. Here's the extract "Increased network width-With IP RIP, the largest possible width of your network is 15 hops. When EIGRP is enabled, the largest possible width is 224 hops. Because the EIGRP metric is large enough to support thousands of hops, the only barrier to expanding the network is the transport layer hop counter. Cisco works around this problem by incrementing the transport control field only when an IP packet has traversed 15 routers and the next hop to the destination was learned by way of EIGRP. When a RIP route is being used as the next hop to the destination, the transport control field is incremented as usual."

Can some1 explain what the above means?

Am I right to interpret as follows

"Below 16 hops, even if EIGRP is enabled, the routers learn the shortest route by RIP way, that is hop.

Above 15 hops, (this is where RIP deems the destination unreachable) then the shortest route is learned by EIGRP metrics of bandwidth, delay, reliability, load, and maximum transmission unit."

Am I right or wrong?

Q2) Pls see attached file "EIGRP 16406" page 35 under section "Pacing Packets". Here's an extract "The default configuration for EIGRP is to use up to 50 percent of the available bandwidth, but this can be changed".

Can someone explain what this mean?

So, does it mean we have to modify the default value to something else?

What would be your suggested value?

Q3) Equal and non-equal load balancing and Feasible Successor, all calculated based on metrics bandwidth, delay, reliability, load, and maximum transmission unit. Right?

So the metric bandwidth, is it referring to the bandwidth of the interface on the router? Such as if it is a FE port, it will take the bandwidth as 100Mbps, regardless of whatever external device it is connected to?

Pls kindly and quickly revert.

Tks!

[jaye15394]

The longest match in prefix length is used first when comparing routes(i.e. a route 10.1.1.0/24 is better than (10.0.0.0/8). This is true when comparing all routing protcols...longest match wins!

If both EIGRP and RIP are enabled on a router and the prefix length is the same, EIGRP will win everytime. It is based on administrative distance EIGRP has an AD of 90 while rip has AD of 120. Lower is better.

If you are comparing routes to the same destination within the same protocol, the best metric wins. This is HOP COUNT for RIP and Bandwidth/Delay (by default) for EIGRP.

Use this site for a beter understanding of route selection.

http://www.cisco.com/en/US/tech/tk365/technologies_tech_note09186a0080094823.shtml

This doc also states max hop count for EIGRP is 220, not 224.

http://www.cisco.com/warp/public/103/eigrp-toc.html

Please rate if helpful.

[rachelhor]

Hi jaye,

But wot about the 50% bandwidth utilisation by EIGRP?

Can explain and advice?

tks!

[pankaj_goyal]

Hello

As for your first query:

RIP uses maximum hop count of 15. That means it can transverse 15 hops and the 16th hop and onwards will be deemed unreachable. The transport control field hop counter is incresed by 1 after every hop. This can go upto 15 hops.

With EIGRP this is not the case. EIGRP uses 'bandwidth' and 'delay' as default metrics. So it can transverse distances spanning many hops (thousands). But the maximum number of hops can only be set to 255. This puts a limitation to the distance it can travel. So to overcome this limitation, what is done is the hop count in the transport control field is incresed by 1 only after EIGRP hops 15 routers i.e. 15 hops by EIGRP are seen as 1 hop to tcp hop counter. This way EIGRP can travel way more than 255 hops.

Also if EIGRP is enabled, router calculates path metrics using only EIGRP no matter how many hops. This is because the 'administrative distance' of EIGRP is lesser than RIP's.

For the second question:

Many routing protocols use 100% bandwidth to advertise their routing tables. This means that the full bandwidth of the link is used by routing protocols to converge. This causes congestion. With EIGRP, only 50% of total bandwidth is utilized for sending routing table updates. This way efficiency in incresed.

You can change this value to any percentage you like but incresing the value will hamper the very cause this is done and decresing it will make routing updates slower. So its better to use the default value.

For the third question:

No. It is not the interface bandwidth that is takes but the link bandwidth. The bandwidth of the link that is connecting two devices is used to calculate the bandwidth metric.

HTH

Regards

Pankaj

[pankaj_goyal]

Sorry, i made a mistake in the previous post. Bandwidth is deduced from the interface type while Delay is associated with the media type connecting two devices.

Both bandwidth and delay can be modified using bandwidth and delay commands.

This is important because by default the bandwidth and delay of serial links is 1544 and 20000 micro seconds respectively. This can cause problems if your are using a high speed link like T1, which has a larger bandwidth, on the serial interface.

[rachelhor]

Pankaj> For your explanation on the bandwidth, can you send me or direct me to the link with some documented proofs so that I can confidently convince my customers & bosses?

Also, to anyone out there in the forum:-

Q4) Can we and should we change the default timer values on the EIGRP to make faster convergence or make it more efficient?

Q5) What is the maximum number of routers that EIGRP can efficiently support? If no benchmark or rule of thumb, perhaps from your experience?

Q6) What are the disadvantages for tuning default timers for any/all routing protocols (EIGRP, OSPF, RIP)? For example if the network consists of more than 100 routers?

Q7) Is it true that it is better to have load balancing (equal and non-equal cost) than having feasible successor in terms of faster convergence and efficiency?

Q8) I need an unbiased and fair opinion.

Comparing OSPF with default timers, OSPF with tuned timers and EIGRP with no feasible successors, what is your opinion?

Pls revert.

Tks!

[ruwhite]

"Can we and should we change the default timer values on the EIGRP to make faster convergence or make it more efficient?"

Please note there are two problems in network convergence: detecting a link or neighbor failure, and then converging around the link or neighbor failure. Faster timers only helps with the detection part, and then only when the link media doesn't provide fast down detection. For instance, a point-to-point SONET link is going to signal down faster than any routing protocol timer, no matter how fast you set it.

"What is the maximum number of routers that EIGRP can efficiently support? If no benchmark or rule of thumb, perhaps from your experience?"

How many balloons fit in a bag? It all depends. With good route summarization, and good network design, there's no real limit. I know of networks running 2000+ routers in a single EIGRP AS. I don't (personally) know of any OSPF networks that large which are pure OSPF--they are all generally broken up with a BGP core, or some such.

"What are the disadvantages for tuning default timers for any/all routing protocols (EIGRP, OSPF, RIP)? For example if the network consists of more than 100 routers?"

Not much, really, other than the work that goes into tuning them.

"Is it true that it is better to have load balancing (equal and non-equal cost) than having feasible successor in terms of faster convergence and efficiency?"

I'm not certain what you mean by "efficiency." In terms of convergence, having two equal cost routes in the local rib is always going to be faster than any other convergence, no matter what the protocol in question. If the link down is detected before the routing protocol neighbor down is detected, and you have multiple equal cost paths, the routing protocol won't even participate in the convergence--until after the tables are already adjusted.

"Comparing OSPF with default timers, OSPF with tuned timers and EIGRP with no feasible successors, what is your opinion?"

This isn't comparing apples to apples, and it depends a great deal on the network design. For instance, in the case of EIGRP without feasible successors, what is the query range? What sorts of processors, and with what sort of load? It all matters.

The order I would say, in general, with caveats, is:

Group 1, generally < 10ms:

1. Equal cost load sharing, any protocol.

Group 2, generally < 50ms:

2. EIGRP with feasible successors.

Group 3, generally < 100ms

3. OSPF within a single flooding domain with highly tuned timers.

4. EIGRP with a short query range (three hops or under), good processors and links, etc.

Group 4, varies widely:

5. OSPF with standard timers.

5. EIGRP with a longer query range, poor links, poor processor, etc.

Network design matters more than protocol.

:-)

Russ