IRC Logs

Network: freenode Channel: #cisco

	Enter your search terms Submit search form
	Web www.irclog.org

[21:33]<gndyvx>well even if you send a supernet downstream longest match will fix your woes
[21:34]<p1nw>reducing the size of the routing table
[21:35]<2zrgvj-l2>autosummarization is done a classful basis.
[21:35]<gndyvx>yeah
[21:35]<gndyvx>no auto.. anywhere, anytime
[21:35]<2zrgvj-l2>just use the network statement with the mask of the summary you want (and are entitled to).
[21:36]<gndyvx>and a null route
[21:36]<gndyvx>network statments require the route to be in the routing table
[21:36]<2zrgvj-l2>yeah.
[21:36]<jygc>f3ew: Reducing how much ?
[21:37]<gndyvx>and then filter the prefix from going downstream and filter longer match prefixes from going upstream
[21:37]<p1nw>oink depends
[21:38]<jygc>f3ew: What's it doing ? Auto-aggregate subnets when it's needed ?
[21:38]<p1nw>See http://www.cidr-report.org/
[21:38]<gndyvx>no such thing as auto-aggrate
[21:38]<gndyvx>unless you own the classful network and thats your boundry
[21:39]<jygc>Yes. I was thinking of 180k -> 80k cidr report ;-)
[21:39]<gndyvx>and you trust something auto (i thought the whole point of BGP was the lack of trust)
[21:39]<2zrgvj-l2>it'll never happen
[21:39]<p1nw>my stupid ISP announces /24s
[21:39]<jygc>err, 120k
[21:39]<2zrgvj-l2>people advertise a lot of those /24s for traffic engineering
[21:39]<p1nw>oink 180
[21:39]<p1nw>120 was 4 years ago
[21:40]<jygc>f3ew: Yes, 120k aggregated
[21:40]<jygc>122 actually.
[21:40]<p1nw>20K new routes added every year
[21:40]<p1nw>most of them /24s
[21:40]<gndyvx>sure
[21:40]<gndyvx>i don't think it's a bad thing
[21:40]<gndyvx>memory is getting cheaper
[21:41]<p1nw>nemith not aggregating where possible?
[21:41]<gndyvx>well when possible
[21:41]<gndyvx>but more and more it is not being possible
[21:41]<jygc>nemith: Well, I don't have much ideas of the impact auto-aggregation can have, actually.
[21:41]<p1nw>If in doubt, aggregate
[21:41]<gndyvx>oink: auto-summary will do it at the classful boundry
[21:41]<gndyvx>classful boundries don't really exist anymore
[21:42]<gndyvx>f3ew: aggregation breaks a multi-homed client
[21:43]<p1nw>If in doubt, aggregate <===
[21:43]<jygc>Right.
[21:43]<gndyvx>know what the fuck you are doing before doing it
[21:43]<gndyvx>my theory
[21:43]<p1nw>If you are administering a network where you have multi-homed clients, yoyu better have the clue needed to run it correctly
[21:44]<p1nw>exactly
[21:44]<p1nw>When in doubt, use brute force
[21:44]<gndyvx>but if you have doubts.. don't fuck with it
[21:44]<gndyvx>:)
[21:44]<jygc>nemith: Actually, memory is probably getting cheaper, but TCAMs.. :-)
[21:45]<gndyvx>sure sure
[21:45]<p1nw>fuckity
[21:45]<jygc>I wonder when we'll reach the 239k limit
[21:45]<p1nw>250K prefixes
[21:45]<gndyvx>as far as cisco is concern
[21:45]<gndyvx>TCAM are large
[21:45]<gndyvx>(ipv6 takes two tcam entries though)
[21:45]<p1nw>or thereabouts
[21:47]<2zrgvj-l2>but no real multihoming solution for the average man in v6
[21:47]<gndyvx>congrats, you have wasted 3 years and 6 months
[21:47]<2zrgvj-l2>hahahahaaa
[21:47]<vz2ufzzzsz>lol
[21:47]<gndyvx>:P
[21:47]<jygc>I'm quite sure they put this limits low so people upgrades their hardware more often
[21:47]<gndyvx>just kiddin
[21:47]<gndyvx>i like reading the cidr-report myself
[21:48]<gndyvx>oink: yeah.. and the goverment is planting mindcontrol chips into my water source
[21:48]<[mzvzyw]>i like reading Sports Illustrated
[21:48]<gjsw>nemith, thanks for the sanity check, your right as usual.
[21:49]<jygc>nemith: exactely. ;-)
[21:49]<gjsw>longest prefix will trump my over-summarised advertisement
[21:50]<p1nw>nosx longest prefix will always win
[21:50]<p1nw>most exact match
[21:50]<p1nw>AS AS Name Current Wthdw Aggte Annce Redctn %
[21:50]<p1nw> Routing Table 245342 217546 22005 49801 195541 79.70%
[21:51]<p1nw>at least a 50% reduction should be feasible
[21:51]<gjsw>is there a good site on bgp etiquette i should check?
[21:51]<2yw>cymru ?
[21:53]<p1nw>nosx the NOG mailing lists, Cymru
[21:53]<ald> <f3ew> fuckity ??
[21:53]<ald>such language
[21:53]<mrxmjum>got 100% in my last exam :D
[21:53]<gndyvx>f3ew: course my company is a top offender
[21:53]<p1nw>nosx the basic rules are: don't advertise anything which isn't assigned to you, or your customers (after verification)
[21:54]<p1nw>and don't bloat the routing table more than you have to
[21:54]<p1nw>cpm you haven't heard me dealing with sopam yet
[21:54]<[mzvzyw]>Ok, VTP is being evil.
[21:55]<[mzvzyw]>9 switches, hub and spoke arrangement
[21:55]<[mzvzyw]>hub is a 5500, running as VTP server, v1
[21:55]<[mzvzyw]>vlan exists and is active, pruning disabled
[21:55]<gjsw>f3ew, those are obvious, im more interested in should i manually summarise everything, what sort of rules should i apply as to prevent a downstream user from advertising a /8 that doesnt belong to them, etc.
[21:55]<[mzvzyw]>one of the spokes is a 3548XL, trunked to the 5500 at the center
[21:56]<[mzvzyw]>vtp client mode
[21:57]<[mzvzyw]>all the vlans show up on the 3548XL, but when I create something on this one particular vlan, 600, it can't ping anything else on that vlan on the other switches
[21:57]<[mzvzyw]>all the other vlans seem to be working though
[21:57]<p1nw>nosx filters?
[21:57]<p1nw>nosx only allow whitelisted advertisments
[21:58]<p1nw>block everything except that which is allowed
[21:59]<vz2ufzzzsz>heh, i remember that issue
[22:00]<[mzvzyw]>tabularasa: This is actually a different switch than was having the issue last time
[22:00]<vz2ufzzzsz>ahhh

Back to #cisco
Or go to some related logs:
#debian
#gentoo
#gentoo
#gentoo
#debian