View Feed
Coffee Room
Discuss anything here - everything that you wish to discuss with fellow engineers.
12889 Members
Join this group to post and comment.
durga ch
durga ch • Aug 13, 2012


Ok, the same have been put in my blog as well (given the popularity of my blog (0)), I though double-posting here might help more people . Here are some quick notes what i joted down for EIGRP. They in no way cover the details and indept topics, but they do cover all the topics what one needs to know.

1.configure 'network x.x.x.x' command to enable eigrp on the interface and to advertise the network as eigrp network.

2.If for specific reasons , neighbourhsip is not required on these iterfaces, use 'passive-interface fa0/1' command.Passive-interface command doesnot sen or recieve EIGRP multicasts onthe interface fa0/1, but would advertise the network x.x.x.x via EIGRP
don't configure the interface as eigrp neighbor with 'network x.x.x.x' command, rather redistribute the interface into eigrp as a connected interface for advertising the network x.x.x.xvia eigrp.

3.'Passive-interface fa0/1' puts the interface fa0/1 into passive mode, whereas 'passive-interface default' puts all the interfaces into passive mode.Interfaces can be made active by using' no passive-interface fa0/2'.

4.neighbors can be configured statically as well by using the command 'neighbor x.x.x.x ser0/1/0'

5.Static configuration of neighbors stops all multicasting advertisement on the interface. Hence icncase of multiple PVCs over single interface, individual static neighbourhsips needs to beconfigured for all potential neighbors.

1.each packet gets authenticated when eigrp authentication is enabled.when the md5 keys dont match authentication fails and thats when neighborships dont form. cars keys, a key is attached to a key-chain . The only difference between car keys and eigrp keys are: eigrp keys can have a duration of life. Once out of duration, the key can nolonger be valid. Key chain, key and key string can be defined with commands - 'key chain name', 'key number' and 'key-string' value.

3.once key-chain, key and key-string are defined, MD5 authentication needs to be enabled for EIGRP using the command - "ip authentication mode eigrp asn md5' and the key to be
used is defined with command 'ip authentication key-chain eigrp asn name-of-chain'.Failing to configure one command leads to authentication failure.

4.While sending packets, lowest valid key-number is used, while recieving packets MD5 digest is checked for all currently valid key

5.Potential EIGRP neighbours need to be in timesync though keychain and hey number need not match, but key-string has to match.

Topology Table:
1.intial seeding via- connected, static neighbours and redistributed routes

2.topology table updation via Update and Query Packet.Update packet has details of prefix , prefix-lenght, metric and non-metric details

3.interfaces whose IP fall withint the network x.x.x.x range, are seeded in topology table as connected routes along with information such as bandwidth, delay, load and reliability details.

4.Update messages containing information about prefix, prefix lenght, load, delay, BW and MTU size are sent.

5.The recieving router, compares the BW in update packet to the BW of the link on which it recieved the packet and chooses the lesser bw for metric. The delay gets summed up.Also
higher load, lesser relaibilty and lesser MTU. Thus EIGRP metric is a combination of minimum BW and cumulative delay.

6.Initial updates tend to be fuller updates once all topology information matches no further updates are sent. Only when a change occurs in the topology, updates are triggered. Also,
updates dont include all prefixes, they only include best possible routes.

7.During neughbourship re-establishement after a flap, full updates are exchanged.Also, EIGRP uses Split horizon which decides which prefixes should be updated, where Split horizons makes sure that advertising a learnt prefix from s0/0/0 back onto s0/0/0 is illogical.

8.EIGRP sends unicast updates on point to point links and multicast updates to on multiaccess domains while expecting unicast Ack from each of the recipient. The process is administered by RTP.
in WAN networks (such as FR)
a.(neighbor establishment)In partial networks, such as hub-and-spoke, EIGRP cannot be established between 2 spokes. reason being, the hellos recieved by hub from spoke1 will not beforwarded to spoke2 at any cost and since there is not PVC between spoke1 and spoke2 , they dont exchange hello's directly as well.So Spoke1 & hub ; spoke2 & hub tend to become neighbors.
b.(topology table updates) Incase both spoke1 and spoke2 are reachable via same interface/subinterface , due to split-horizon property of EIGRP, hub would not tend to forward updates w.r.t prefix x.x.x.x learnt from spoke1 via ser0/0/0.1 onto ser0/0/0.1 towards spoke2. Thus spoke1 and spoke2 are unaware of each others networks as well.
c.In NBMA networks, since multicasts are not allowed, unicast EIGRP updates are sent to each neighbour which might end in excessive BW comsumption.This can be controlled by
allocating a specific %of BW for EIGRP process , thus enabling IOS to queue the updates and delay the process .

Routing Table:
1.The routes to be imparted into the routing table are deduced from topology table. EIGRP doesnot perform SPF algorithms rather depend on paramters such as Feasible distance and
reported distance.EIGRP topology table hols only best learnt complete routes unlike OSPF which stores all subnets, links in database table and runs complex algorithms to find shortest path.

2.FD is used to determine the best path where as RD is used to determine the next-best path or fall back path.

3.The route with lowest possible FD for a given prefix is inserted into IP Routing table

4.Metric = 256 * ((10^7/lowest BW on entire route)+ cumulative delay)

5.The bandwidth and delay subinterface commands can influence the bandwidth and delay of updates going via that specifc interface. wAN interfaces, where a single WAN link to FR cloud, might support multiple VCs, oversubscription og BW is done where , the BW allocated would typically not be totalbw/Numberof VCs rather the ratio cumulative allocated BW\total BW available is app 2:1

7.Apart from EIGRP calculations, BW is used as an input paramter for QoS calculations, CIR rats determination. Hence best possible metric to be tweaked in delay.

When offset-lists are enabled, a defined value is added to the existing metrics of a given prefix.An offset can be defined with command - 'offset 1 in 130 ser0/0/0', where 1 is
accordingly- access-list which points to prefixes, 130 is the offset value , 'in' is direction of applciation of offset which defines the direction in which the updates are send and ser0/0/0 is the application of offset value.

9.EIGRP metrics are manipulated for better convergence as well.

EIGRP convergance:

1.RD as calculated by router helps router in determining if a specific route can be considered as loop-free back-up route.

2.The route with smallest metric for a given prefix is known as a successor and the next-hop in such route is known as a successor.Any other best routes are held in topology table for later use.These routes are known as feasible successors.

3.Feasibility condition:
If non-succesors RD is less than FD then it is eligible to become a Feasible successor.This condition also implies that, A's metric to subnet x is 10000 via Q. If P's metric to x is 9000, it is obvious to A that P is not learning the prefix via A else, P's metrics to x would be >10000. Thus any RD >10000 is seen with skeptism by A.

4.When EIGRP succesors goes out-of-sight from ip routing table, the feasible succesor ( if any ) is inserted into the IP routing table, in case no feasible succesor is present 'QUERY'
packets are sent to know if any of the existing neighbours are aware of a specific prefix.

5.For these prefixes, for which neither a successor nor a feasible succesor exists in topology table,EIGRP sets the prefix in active state (going-to-active) and send a 'QUERY' packet
asking for a loop-free route for the prefix.

6.unless and until all the 'QUERY' packets are 'REPLY'ied, the router cannot decide which REPLY is the best

7.stuck-in-active state:
When a router A sends a QUERY for prefix X to a neighbor B and B too is in active for this route, and is in process of sending and recieving QUERY/REPLY, it shall not reply back to A.
Router A is now stuck waiting for reply from B. This time duration for which A can wait for B to reply is known as active timer.Thus Router A is SIA for route x.

8.To reduce the scope of QUERY packets, end nodes (such as branch nodes) can be configured as stubs apart from route filtering techniques which can be used to filter which prefixes will be learnt from what neighbours.

9.Stubs:Stub routers dont advertise routes learnt from one neighbor to another neighbour and dont entertain QUERY packets as well.They do tend to form neighbourship, but thats
where it all ends.

10.Routes advertised as summaries too reduce the number of QUERY packets. For example a summaried route for /16 subnet X exists in routing table of A. when A recieves a QUERY
regarding /24 of X , it wouls simply reply 'NO' .

11.Unequal metric load balancing is attained by using all the unequal metric paths by adjusting the value of variance and as well maximum number of paths to be entered in Routing table can as well be set. Variance is an interger(i) multipled to FD (say via A). A prefix metric learnt via B when < i * FD , can be considered as a possible route to be inserted into IP Routing table.The number of such paths which will be uncluded in routing table is limited by maximum-path paramter.

Share this content on your social channels -