Priscilla Oppenheimer wrote: > > ""Tim Champion"" wrote in message > news:[EMAIL PROTECTED] > > When calculating the metric of an IGRP route (with > non-default 'K' values) > > which load and reliability values does one use? Do you use > the highest, > > lowest or average value for the entire route?
Load and reliability are cumulative as far as I can tell. That is, they reflect the load and reliability for the entire path. That's what Doyle says and I now believe him. :-) You can test it without even messing with the K values. To see the reported load and reliability, use a protcol anzlyer, or, even easier: do a show ip route with the network of choice as a parameter. The info is shared and saved even, if it's not used to calculate the composite metric. > > > > Also if anyone could point me to a document on the above it > would be > > appreciated. > > > > > > Many thanks in advance. I did some testing. It was impossible to come up with a testing methodology that would give a conclusive result, especially considering the limitations of my lab, and the limitations of the protcols, but I'm pretty sure that both load and reliability are cumulative. Well, for reliabiity, the error rate is cummulative really. And load does seem to be somewhat smart about different bandwidths. It can't be too smart since it only knows the lowest bandwdith for the path, but it seems to take this into account along with the bandwidth for its interfaces, which is necessary for a cumulative load to mean anything. Needless, to say, the real answer is don't even bother with reliabilty and load as IGRP or EIGRP metrics. With a 90-second update timer (IGRP) and nonperiodic, partial, and bounded updates for EIGRP, the protocol can't possibly keep accurate track of the sampling that interfaces do every 5 seconds to determine the load and reliability. It's kind of weird that Cisco let these two opposing goals end up in the protocol: * Low bandwidth usage by the protocol (infrequent updates) * Path selection based on quickly-changing load and reliabity metrics Well, probably nobody cares (and rightly so! ;-) But it was bothering me that the original poster never got a good answer. Priscilla > > Remember this oldie but goodie question? Most of the answers > were links to papers that don't answer the question. I answered > the question by saying that load is the highest load on any > segment in the path and that reliability is the worst > reliability on any segment in the path. > > However, I noticed that Doyle says that load and reliability > are cummulative. > > For reliability, he says: Reliability reflects total outgoing > error rates of the interfaces along the route. > > For load, he says: Load reflects the total outgoing load of the > interfaces along the route. > > > So, if Doyle is right, I guess we can assume that a router > accepts a route from a "downstream neighbor," looks at > reliability (expressed as a fraction of 255) in the route, > looks at the reliability on the incoming interface (also > expressed as a fraction of 255), and somehow munges those > together to form a reliability figure that it should pass on to > its "upstream neighbors" for that route. Would it multiply the > values?? > > Assuming a router actually cares about this part of the > composite metric, the inverse of this reliabilty result gets > folded into the metrc. (I say inverse because metric is "cost" > and we care about the error rate not the reliability). > > > For load, if Doyle is right, we can assume that a router > accepts a route from a downstream neighbor, looks at load > (expressed as a fraction of 255), looks at load for the > incoming interface (also expressed as a fraction of 255), and > somehow munges those together to form a load figure that it > should pass to its "upstream neighbors." Does it add them?? > Wouldn't that be adding apples and oranges? 255/255 load on a > T1 is different from 255/255 load on Fast Ethernet. Of course, > the router does have the bandwidth for the route too, so it > could use that for calculating a cummulative load too. > > > So, I think my answer makes life much easier for the routers. > Just report whichever is worse, the one in the incoming route > or the one for your incoming interface. The result is that the > load for a route reflects the highest load of any segment and > the reliability reflects the worst reliability for any segment. > It's not quite as granular, but I think it would be fine as a > metric. I really did think it worked that way, though I can't > find any documentation that says this. But I think a Cisco > person told me this..... > > > So I tried to do some testing, but of course, load and > reliability are so variable, and IGRP and EIGRP are so not > variable, that I didn't have much luck determining the truth. > Plus I managed to put routes in holddown by increasing the load > too much! > > With IGRP updates only being sent every 90 seconds, they > obviously don't take into account real-time load and > reliability anyway. And with EIGRP, there are no updates unless > there are changes. So the inclusion of load and reliability are > silly, but I still wonder what they mean if they are used. > > > If I had enough sniffers I could have tested IGRP. A 90-second > update timer isn't too painful. Even without the metric weights > command, the info is passed in the updates. I would need a > couple sniffers to see how the values change compared to the > router interface values. > > By the way, even when reliability and load aren't used in the > composite, you can still see their value for a route in the > routing table if you do "show ip route . So testing my > theory is possible, but not practical in my limited lab network. > > > So, has anyone gotten this far in this long-winded message? :-) > Does anyone know how load and reliability are really > calculated? Would anyone like to take on the project to test > it? Is it the worst or cummulative? > > Please don't send me to a URL that doesn't really answer the > question. The ones that cover the weighting (K values) don't > answer the question. The one that talks about FTP and the > question of how often load is calculated also doesn't answer > the question of whether load is cummulative or the worst case > when calculating the metric for a route. > > Thanks for listening and thanks in advance for anyone who would > like to help me unravel the mystery!? (Or should we just > believe Doyle! That usually works!) > > Priscilla > > > > Message Posted at: http://www.groupstudy.com/form/read.php?f=7&i=66578&t=66522 -------------------------------------------------- FAQ, list archives, and subscription info: http://www.groupstudy.com/list/cisco.html Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]
