Full IBGP deployment experience

Hi guys,

I'm wondering if any of you deployed a full IBGP on your network - that is not only on some edges, but a full one? do you have any tough experiences regarding this? concerning designs, RR placement etc?

any input appreciated, thank you.

Find more posts tagged with

Comments

kryolla

why would you want to run a full iBGP when IGP is faster to converge and easier to set-up/maintain. You have to think of it from a design perspective does it make sense. Just because you can do it doesnt mean you should. This isnt directed at you but I find it funny how people just learning new technology tries to come up with new ideas and wonder why it doesnt work the way it is suppose to. Like what Brian from IE keeps on saying find out why it was created and what issues does it solve and the technology will make sense. Its good to try and tweak a situation to make it more complicated to see if you understand it but you really have to know the design purpose of the technology. Just my rant for today.

EDIT my response was for enterprise and not service provider, my apologies

adriansizemore

kryolla wrote: »

why would you want to run a full iBGP when IGP is faster to converge and easier to set-up/maintain. You have to think of it from a design perspective does it make sense. Just because you can do it doesnt mean you should. This isnt directed at you but I find it funny how people just learning new technology tries to come up with new ideas and wonder why it doesnt work the way it is suppose to. Like what Brain from IE keeps on saying find out why it was created and what issues does it solve and the technology will make sense. Its good to try and tweak a situation to make it more complicated to see if you understand it but you really have to know the design purpose of the technology. Just my rant for today.

I am not sure about your assessment. Certainly, under certain network infrastructures, an IGP (non distance vector) would be useful, but in a large ISP environment, I lean towards BGP, and even MBPG.

I think some follow up questions are needed for the original question The first, and most obvious, how big of a network are you talking about? What is the topology? etc... I maintain a large public internet backbone, and a large private backbone at work. Both networks use BGP for customer data, but also use TE-MPLS and MPLS/VPNs, which as we all know rely on a full IGP backbone.

rossonieri#1,
I have to assume you are speaking of modeling a large ISP environment when you mention iBGP mesh and route reflectors; in this case, I have not set one up yet and looking forward to modeling it in my lab. As for problems (ISP size), most i've seen are timers from SONET/SDH BIP errors to BGP, which I hardly doubt you will experince in your lab.

As for the lab, we can compare notes. I have my network lab (GNS3) up and running, and will be working on a full iBGP mesh when I am done with IGP between PE-CE connections, to include backbone and non backbone connections.

rossonieri#1

thanks for the input kryolla & adrian

@ kryolla,

with all due respect to mr. bryant - he's a great tutor, i just like to brainstorming with you guys - if any of you have deployed a full IBGP network, and like to share some experiences with me, more like pushing IBGP to the limits.

@ adrian,
thanks for sharing. how many core do you have inside your backbone?
do you guys also split your core into different customer segment? you know, retail (public segment?), corporate (private?) etc?

keenon

If you have need of a full iBGP deployment you should be thinking on confederations (with/and\or route-reflectors included) but if your just labbing then have at it once you get past 4 routers you start rethinking the idea

rossonieri#1 wrote: »

thanks for the input kryolla & adrian

@ kryolla,

with all due respect to mr. bryant - he's a great tutor, i just like to brainstorming with you guys - if any of you have deployed a full IBGP network, and like to share some experiences with me, more like pushing IBGP to the limits.

@ adrian,
thanks for sharing. how many core do you have inside your backbone?
do you guys also split your core into different customer segment? you know, retail (public segment?), corporate (private?) etc?

rossonieri#1

hi keenon,

hey how are you? where have you been? long time not see you hanging around TE lately, busy? anyway, thanks for sharing.

confederations, yup - that was my first picture

about the lab? i'm trying to do with at least six routers using every possible design so they can do all IBGP net. but how come every design that comes to my mind was in triangle shape? i'm trying rectangle like design 3 routers in a row x 2 rows

hey, when will you take the lab keenon? ed, he seems at this june - if i'm not mistaken. anyway - good luck on the lab preparation

keenon

actually its aug 1.. thats the due date.. starting to lab a bit harder than the roundabout way i have been doing. i've been showing up a little more randomly than i have been over the past 1 or 2

i used a 6 router lab config for route reflectors and confederation using dynamips.

autostart = False
[localhost:7200]

3725
image = \Program Files\Dynamips\images\c3725-AD.bin
# On Linux / Unix use forward slashes:
# image = /opt/7200-images/c7200-jk9o3s-mz.124-7a.image
ram = 128
mmap = True
sparsemem = True
ghostios = True
idlepc = 0x60bec7a8
######################################################################
ROUTER SW1
model = 3725
autostart = False
slot1 = NM-16ESW
f1/0 = SW3 f1/0
f1/1 = SW3 f1/1
f1/14 = SW2 F1/14
f1/15 = SW2 F1/15
f1/7 = SW4 f1/7
f1/8 = SW4 f1/8

ROUTER SW2
model = 3725
autostart = False
slot1 = NM-16ESW
f1/0 = SW4 f1/0
f1/1 = SW4 f1/1
f1/7 = SW3 f1/7
f1/8 = SW3 f1/8

ROUTER R3
model = 3725
autostart = False
f0/0 = SW1 F1/2

######################################################################
[localhost:7201] # Talk to the 2nd dynamips process on port 7201
udp = 11000

3725
image = \Program Files\Dynamips\images\c3725-AD.bin
# On Linux / Unix use forward slashes:
# image = /opt/7200-images/c7200-jk9o3s-mz.124-7a.image
ram = 128
mmap = True
sparsemem = True
ghostios = True
idlepc = 0x60bec7a8
#######################################################################

ROUTER SW3
model = 3725
autostart = False
slot1 = NM-16ESW
f1/14 = SW4 f1/14
f1/15 = SW4 f1/15

ROUTER SW4
model = 3725
autostart = False
slot1 = NM-16ESW

ROUTER R5
model = 3725
autostart = False
f0/0 = SW4 F1/2

rossonieri#1

hi keenon,

that 6 routers looks like this?

           ==== sw2 ====
r3 --- sw1        ||          sw3 --- r5
           ==== sw4 ====

i was thinking to lab this set up :

r1 --- r2 --- r3
|        |       |
r4 --- r5 --- r6

keenon

rossonieri#1 wrote: »
hi keenon,

that 6 routers looks like this?
           ==== sw2 ====
r3 --- sw1        ||          sw3 --- r5
           ==== sw4 ====
i was thinking to lab this set up :
r1 --- r2 --- r3
|        |       |
r4 --- r5 --- r6

it is and allows with a few changes to do some other configurations... i do have one lab i run for full mpls lab.. that has 10 routers if you want. that i have used for standard, TE, CSC, ATOM, central services

rossonieri#1

keenon,

i do have one lab i run for full mpls lab.. that has 10 routers if you want. that i have used for standard, TE, CSC, ATOM, central services

wow, thanks - i would love too, that would be great

considering random scenario possibilities

but wait - 10 routers, do you play that on a single machine or do you split them? GNS or pure dynamips?

keenon

rossonieri#1 wrote: »

keenon,

wow, thanks - i would love too, that would be great
considering random scenario possibilities

but wait - 10 routers, do you play that on a single machine or do you split them? GNS or pure dynamips?

1 machine, 2 dynamips server processes. I run it on 2 different laptops 1 Duo core 1.6ghz with 2gb ram thats when i'm at home and the other is Duo core 2.16ghz with 4gb ram at work. i run pure dynamips not a gui type

rossonieri#1

1 machine, 2 dynamips server processes.

wow, i'd never thought that this *multi-core CPU* thing has great impact for dynamips

my bad, i only have a couple of old p4 machines at home - so, you'll note a very drastic performance degradation when running more than 4 to 6 routers, even those idle-pc values has been setup.

hey, thanks for the map btw