>This is a very interesting and complex question in my opinion. As I
>understand it the classification of whether a routing protocol utilizes a
>distance-vector or link-state algorithm depends on how that routing protocol
>announces it's routes to the network.
>
>With a distance-vector algorithim each device will include in it's
>announcements the destinations to which it is directly attached (depending
>on configuration) as well as destinations that it has heard about from other
>devices (also depending on configuration and route filtering).
>
>In contrast, in a link-state protocol, a device doesn't provide information
>about destinations it knows how to reach.
There's something lurking here about the reachability of layers in
the external clothing and bra stack.
>Instead, it provides information
>about the topology (no pun intended from the previous phylosophy thread:) of
the network in its immediate vicinity.
To keep this at least PG, I shall refrain from observations about
topology versus bottomology.
>Each device announces link-states of
>it's interfaces (again, depending on configuration) to all devices in the
>network or area. This ensures that all devices within that area/network
>have an identical database or picture of the network and can thus make
>routing decisions utilizing the SPF algorithm. As you can see, using a link
>state protocol will eliminate the need for route loop detection as loops
>could not exist if all devices in the network have an identical view of that
>network.
>
>In my mind, though BGP doesn't neatly fall into either category being
>labeled as a path-vector routing protocol, it would generally be considered
>more of a distance-vector algorithm because it doesn't have a consistent
>database of link-states but instead uses a table of routes that were
>announced to it from a BGP peer.
No question that BGP is more like a DV protocol, although it has
definite aspects of source routing with loop prevention. An
alternative protocol for exterior routing, IDPR, consciously was
derived from LS.
>
>EIGRP is along the same lines as far as classification is concerned. Though
>it establishes adjacencies, has low convergence time, and sends
>incremental/partial routing updates, it still uses the DUAL algorithm to
>find the best route from a neighboring routers route table and not a
>consistent link-state database. Cisco may claim this is a hybrid
>link-state/distance-vector protocol, but to me it is still distance-vector.
Cisco marketing may call it "hybrid," but JJ Garcia-Luna, who
developed the DUAL algorithm, definitely thinks of it as an evolved
DV. Several of the key Cisco architects and developers, such as Dino
Farinacci (now Procket) and Dave Katz (now Juniper), can be very
eloquent about the virtues of advanced DV protocols. I _think_ Tony
Li (now Procket) prefers LS, but it's never quite that simple. Some
of the theoretical advantages of a well-designed DV include more
tolerance of errored routing information (e.g., Byzantine corruption)
and less CPU demand.
I like to categorize DV protocols into generation:
1st: Hop count metric, periodic update, dead router detection via
missing updates, loop handling with count-to-infinity, split
horizon, and holddown (e.g., IP RIP, RTMP, IPX RIP)
2nd: Bandwidth/delay metric, periodic and triggered update with some
randomization of update timing, dead router detection via missing
updates, additional loop detection such as monotonically increasing
metric (e.g., IGRP and partially some advanced RIP implementations
that still use hop count)
3rd: Bandwidth/delay metric, hello subprotocol for dead router
detection, updates on changes only, reliable transfer of updates,
advanced loop prevention
Note that classless or classful capability really doesn't have
anything to do with the protocol behavior itself.
>If anything can be considered a hybrid protocol it would be OSPF as it uses
>both link-state databases (intra-area routes) and distance-vector
>advertisements (inter-area routes) when the ABR advertises all networks from
>a non-backbone area into a backbone area. Right Howard??
Correct. The Dijkstra algorithm is used only for non-backbone
intra-area routes.
>
>-Mike Cohen
>
>-----Original Message-----
>From: cisco cabanaboy [mailto:[EMAIL PROTECTED]]
>Sent: Friday, May 26, 2000 10:59 AM
>To: [EMAIL PROTECTED]
>Subject: BGP
>
>
>Is BGP DV, or LS?
___________________________________
UPDATED Posting Guidelines: http://www.groupstudy.com/list/guide.html
FAQ, list archives, and subscription info: http://www.groupstudy.com
Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]
