Chris,
On 03/01/2022 20:23, Christian Hopps wrote:
Peter Psenak <[email protected]> writes:
Chris,
On 03/01/2022 17:18, Christian Hopps wrote:
Peter Psenak <[email protected]> writes:
On 03/01/2022 16:21, Christian Hopps wrote:
On Nov 29, 2021, at 7:39 PM, Les Ginsberg (ginsberg)
<[email protected]> wrote:
Tony –
Let me try one example – see if it helps.
Summarization is used in the network.
But customer identifies a modest number of key nodes where it wants to detect
loss of reachability ASAP. Unfortunately, customer is unable to assign
addresses which are outside of the summary to these nodes.
I think this does in fact capture the problem trying to be solved here, nicely.
not really.
In fact assigning addresses to the nodes in a way that they are part of the
summary is the right thing to do.
No, not if you want more detailed information about specific reachability it's
not. And ....
typically you want to summarize all prefixes inside the area when advertising
outside the area. And you want to know about some of these prefixes when they
are lost to help convergence.
The problem we are trying to solve is to use the summarization but without the
loss of the fast notification of the node down event.
You want more specific information about reachability, but you just want to do
it when the network is stressed and undergoing change.
So the "works now" way of not summarizing these important prefixes has the
state in the network when it's working, so you know adding and removing it is
something the network is already capable of handling.
New signaling that *only* is created when things start failing, tests the
infrastructure at exactly the wrong time.
In 99,99% of cases there will be only single pulse generated when one PE goes
down. That itself is a very rare event itself.
You can say the same thing about all the regular state already carried in the
IGP.
not really. Node failures happens significanly less often than link
flaps. That's the fact that any ISP can confirm.
We can easily limit the number of pulses generated on ABR to a single digit
number to cover the unlikely case of many PEs in area becoming unreachable at
the same time.
If a failing network can handle the extra state, then a functioning stable
network of course can too.
no, that's not what we claim. We want network to be summarized all times and
generate limited number of pulses at any given time to help the network converge
fast in case where single (or very few) PEs in an area go down.
And I'm saying if a prefix is important enough to merit a bunch of new protocol
extensions and state, then it's important enough to simply be left out of the
summarization in the first place.
And then people get what they want, w/o protocol changes/upgrades, and it's
using time tested and hardened IGP code and designs.
let's imagine you have 100k inter-area prefixes in the network - 1k
prefixes per area, 100 areas, 2 ABRs per area. With summarization you
compress to 1 summary per area. You will have 100 summary prefixes
advertised, each by two ABRs, so 200 summary prefixes.
If one PE fails you will have 200 summaries + 2 pulses (2 ABRs per area).
If one PE fails in every area at the same time you will have 200
summaries + 200 pulses.
If we limit the number of pulses per ABR to 10, regardless of how many
PE fail at the same time, you will end up with 200 summaries and 2k pulses.
Without summarization you will have 200k states.
In the unrealistic worst case you will have 200 summaries and 2k pulses.
2,2k is in the order of the magnitude less then 200k. And we solved the
problem of PE down for any realistic case.
Peter
Why doesn't this work?
Thanks,
Chris.
thanks,
Peter
Thanks,
Chris.
_______________________________________________
Lsr mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/lsr
