On 11 Aug 2011, at 22:40, George, Wesley wrote:

> Danny said,
> "Periodic updates of the entire routing table *with much larger and more 
> updates* seems undesirable at best to me, particularly to 'reduce the 
> vulnerability window for replay attacks' to 'days'."
> ---------------------------------------------------------
> I think that this is a small part of a much larger issue that has been 
> bugging me since our meeting in Quebec City.
> After looking at Sriram's estimates around the necessary amount of memory, 
> etc for the different algorithms, I'm bothered by what is being asked of the 
> routing system in support of BGPSec.
> 

Hi All,

Wes' point here is really important I think, and it's a bit of a shame it 
hasn't got more focus. I stood up and mentioned scaling in Prague, then found 
myself making a very similar point again in Québec. As with Wes, I would prefer 
that this is not interpreted as a criticism of the goals of bgpsec, but rather 
is an operational perspective on the assumptions that are made for deployment 
of the protocols being developed in sidr.

The root assumption that we seem to be taking as read throughout such 
discussions is that at the point where routing infrastructure is expected to 
support the computational load that is demanded of it by bgpsec, there will 
have been sufficient growth in both the computational and memory resources that 
are available to these systems. Whilst I appreciate that cryptographic 
validation of the validity of a route can be performed on the edge system(s), I 
think there are two very key concerns here.

- Historically, do we see Internet routing systems matching the rate of general 
computational resource availability? My view on this is that we do not - there 
are plenty of edge systems out there that are relatively modern (and capable of 
routing tens of gigabits of traffic) that do not have a control-plane that is 
as powerful as some of the software devices that existed prior to it - as such, 
this means either this was a lower priority to deliver (which may well be 
true), or that it was not practical to do so. With increasing demands on 
control-plane resources of IP systems (be they Internet edge or otherwise) - 
there is surely some query as to whether having a large amount of memory or 
crypto offload hardware will be a priority over other features. The question 
Wes raised as to whether we, as a working-group, are happy with the assumption 
that this is the case, is very valid. It would seem to me that we are being 
somewhat prescriptive with prioritising what hardware vendors' solutions should 
look like, and what operator's scaling approaches will look like - without 
necessarily explicitly discussing why this is reasonable to do so.

- Are the assumptions that we are making regarding how those autonomous systems 
delivering an "Internet" product set are architected valid? There is some 
assumption (AIUI, please feel free to correct me) that we are looking at 
networks where we have explicit edge devices which can scale to meet the 
validation workload that we're putting on it. This would seem (again, to me) to 
be based around assumption that there are a relatively high number of 
less-dense edge devices. I would be interested whether it's the feeling of the 
WG that my reading of what we're assuming is correct - and if so, whether the 
WG feels that we can be prescriptive like this for how one should architect an 
Internet network.

I'm again very supportive of what problems bgpsec is trying to solve - and do 
not want to put a dampener on any of the excellent efforts that have gone into 
it - but think that we cannot wholly abstract ourselves away from the actual 
deployability of the solutions that are being developed. After all, we will not 
solve the problems that the WG is chartered to solve if the fruits of the 
labours herein are not actually deployed.

I am not sure of exactly how we progress discussing the scalability of the 
solution - but with Sriram's analysis, is it possible to discuss within this 
forum, and perhaps others concerned with the general scaling of IP routing 
elements, why the scaling considerations/assumptions of bgpsec are valid, and 
where a departure is made from the longer term view of scalability that is 
presented in existing IETF documents.

Thanks for reading this - and your consideration!
r.





_______________________________________________
sidr mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/sidr

Reply via email to