The original motivation for developing EIGRP was simply to make IGRP classless. But early in the development the engineers working on the project recalled some academic proposals for a new kind of convergence algorithm and decided to use that algorithm in their extension of IGRP. The result was a protocol that, while retaining some concepts introduced with IGRP such as multiple metrics, protocol domains, and unequal-cost load balancing, is distinctly different from IGRP.
EIGRP is occasionally described as a distance vector protocol that acts like a link-state protocol. To recap the extensive discussion in Chapter 4, "Dynamic Routing Protocols," a distance vector protocol shares everything it knows, but only with directly connected neighbors. Link-state protocols announce information only about their directly connected links, but they share the information with all routers in their routing domain or area.
All the distance vector protocols discussed so far run some variant of the Bellman-Ford (or Ford-Fulkerson) algorithm. These protocols are prone to routing loops and counting to infinity. As a result, they must implement loop-avoidance measures such as split horizon, route poisoning, and hold-down timers. Because each router must run the routing algorithm on received routes before passing those routes along to its neighbors, larger networks might be slow to converge. More important, distance vector protocols advertise routes; the change of a critical link might mean the advertisement of many changed routes.
Compared to distance vector protocols, link-state protocols are far less susceptible to routing loops and bad routing information. The forwarding of link-state packets is not dependent on performing the route calculations first, so large networks might converge faster. And only links or prefixes and their states are advertised, not routes, which means the change of a link will not cause the advertisement of all routes using that link.
Regardless of whether other routing protocols perform route calculations before sending distance vector updates to neighbors or after building a topological database, their common denominator is that they perform the calculations individually. In contrast to the Bellman-Ford algorithms used by most other distance vector protocols, EIGRP uses a system of diffusing computationsroute calculations that are performed in a coordinated fashion among multiple routersto attain fast convergence while remaining loop-free at every instant.
Although EIGRP updates are still vectors of distances transmitted to directly connected neighbors, they are nonperiodic, partial, and bounded. Nonperiodic means that updates are not sent at regular intervals; rather, updates are sent only when a metric or topology change occurs. Partial means that the updates will include only routes that have changed, not every entry in the route table. Bounded means that the updates are sent only to affected routers. These characteristics mean that EIGRP uses much less bandwidth than typical distance vector protocols use. This feature can be especially important on low-bandwidth, high-cost Wide Area Network (WAN) links.
Another concern when routing over low-bandwidth WAN links is the maximum amount of bandwidth used during periods of convergence, when routing traffic is high. By default, EIGRP uses no more than 50 percent of the bandwidth of a link. Later IOS releases allow this percentage to be changed with the command ip bandwidth-percent eigrp.
EIGRP is a classless protocol (that is, each route entry in an update includes a subnet mask). Variable-length subnet masks may be used with EIGRP not only for sub-subnetting as described in Chapter 6, "RIPv2, RIPng, and Classless Routing," but also for address aggregationthe summarization of a group of major network addresses.
EIGRP packets can be authenticated using an MD5 cryptographic checksum. The basics of authentication and MD5 are covered in Chapter 6; an example of configuring EIGRP authentication is included in this chapter.
Finally, a major feature of EIGRP is that it can route not only IP but also IPX and AppleTalk.
——————————————————————————————————
開發EIGRP的早期，由於接受了學術建議使用新的的收斂算法，而且最後決定用這個算法來擴展IGRP。導致的結果是，除了IGRP的複合度量，協議域，非等價負載均衡的概念，其他都和IGRP不同。
EIGRP協議有時也被描述成具有鏈路狀態協議的距離矢量協議。距離矢量協議分享他知道的所有信息，但是範圍僅傳遞給鄰居路由器；鏈路協議僅通告他們的直連鏈路，但是在域內或者區域內共享所有路由信息。
目前爲止，所有距離矢量協議都是基於BF算法或者他的派生。這些協議更易於產生路由環路和計數到無窮大。結果，必須產生環路避免手段，比如水平分割，路由毒性，抑制計時器超時。由於每臺路由在他傳遞給路由給他鄰居前都要先運行路由協議，所以大型網絡收斂很慢。更主要的是，距離矢量協議通告路由時，如果關鍵鏈路發生變化將導致許多產生變化的路由器都發送通告。
和距離矢量比較，鏈路狀態協議更不易受路由環路和錯誤路由信息的影響。鏈路狀態數據包的轉發不是先根據計數路由，所以大型網絡可能收斂快些。而且只有鏈路，前綴和狀態可以被通告，不通告路由，結果是鏈路的改變不會導致所有用那條鏈路的去通告。
其他任何協議，不管是先計算路由表再傳送給鄰居的距離矢量方式還是先建立拓撲數據庫，他們通常都是進行獨立路由運算。然而，EIGRP用了一種不同BF的算法，他用一個稱爲擴散算法的方式，效果就是在多臺路由器上進行整體協調改變從而到達快速收斂並且隨時都保持無環。
雖然EIGRP更新仍然是距離矢量型方式傳輸給直連鄰居。他的特性是非週期，局部，有邊界的。非週期：更新的發出沒有更新間隔，只有當度量和拓撲發生變化時才發送。局部的：只有涉及到的路由條目纔會發送更改。有邊界的：更新只發送給受影響的路由器。這些特性意味着他只需要更少的帶寬相對於傳統的距離矢量協議。這些特性十分重要在於低帶寬，高價格的WAN鏈路上。
需要注意的，當路由流量比較高，而卻在鏈路帶寬比較低的WAN鏈路上時，路由期間需要用到最大的帶寬去收斂。默認這個EIGRP用的鏈路帶寬不超過50%，在之後的IOS，可以用命令ip bandwidth-percent eigrp來修改這個百分比。
EIGRP是一個無類協議（一個更新中的每個路由條目都一個子網掩碼）。可變長子網掩碼用在子網掩碼就想第六章的效果，而且可以用於地址聚合來彙總一組主類網絡地址。
EIGRP包可以通過MD5加密校驗。這個基本驗證和MD5方式在第六章講過了。有一個EIGRP驗證配置包含在這個章節中。
最後，一個主要的特性是EIGRP不僅可以路由IP數據包，也可以路由IPX和AppleTalk數據包。
————————————————