Hi Huaimo,

> Introducing centralized feature into IGP will break IGP's distributed
nature

That clearly proves that word "centralized" has been significantly
overloaded here.  To many indeed "centralized" means a controller (like
OpenFlow or SDN) and that such device added to a network is to push
information - typically 1RU linux blade -  here optimized flooding graph.
But this never was the plan with this proposal from its start ie. -00
version.

Centralized means that optimized flooding graph comes from single redundant
node.

Leader election happens automatically and procedures for that are to be
vastly similar to today's DR or DIS election. So with this in mind one may
observe that both OSPF and ISIS are pretty centralized on multiaccess
networks today :)

To your point of multi-vendor networks true - and that is precisely why
upgrade network wide to a release containing consistent algorithm from more
then a single vendor (and even for single vendor) is practically a very
time consuming and difficult process.

Btw I don't think there is any problem here ... The text added to -05
version allows very seamless choice of centralized vs distributed topology
computation by signalling either zero or non zero value in the added to
version -05 area leader sub-tlv.

In other words I don't see any problem or room for debate .. adopting and
implementing -05 allows use of centralized or distributed optimal flooding
computation at the operator's discretion.

Thx,
R.

On Mon, Aug 27, 2018 at 5:10 PM, Huaimo Chen <huaimo.c...@huawei.com> wrote:

> >> I think distributed is more practical too.
> >I would appreciate more detailed insights as to why you (and others) feel
> this way.  It is not at all obvious to me.
> IGP is distributed in nature. The distributed computation of flooding
> topology like distributed SPF will keep IGP still distributed in nature.
> Introducing centralized feature into IGP will break IGP's distributed
> nature, which may cause some issues/problems.
>
> >> For computing routes, we have been using distributed SPF on every node
> for many years.
> >True, but that algorithm is (and was) very well known and a fixed
> algorithm that would clearly solve the problem at the time. If we were in a
> similar situation, where we were ready to set an algorithm in >concrete, I
> might well agree, but it’s quite clear that we are NOT at that point yet.
> We will need to experiment and modify algorithms, and as discussed, that’s
> easier with a centralized approach.
> After flooding reduction is deployed in an operational (ISP) network, will
> we be allowed to do experiments on their network?
> After an algorithm is determined/selected, will it be changed to another
> algorithm in a short time?
>
> >> In fact, we may not need to run the exact algorithm on every node. As
> long as the algorithms running on different nodes generate the same result,
> that would work.
> >Insuring a globally consistent result without running the exact same
> algorithm on the exact same data will be quite a trick.  Debugging
> distributed problems at scale is already a hard problem.  Having >different
> algorithms in different locations would add another order of magnitude in
> difficulty.  No thank you.
> In some existing networks, some nodes run IGPs from one vendor, some other
> nodes run IGPs from another vendor, and so on. Some may use normal SPF,
> some others may use incremental SPF. It seems that we have had these cases
> for many years.
> >Tony
>
> Best Regards,
> Huaimo
> _______________________________________________
> Lsr mailing list
> Lsr@ietf.org
> https://www.ietf.org/mailman/listinfo/lsr
>
_______________________________________________
Lsr mailing list
Lsr@ietf.org
https://www.ietf.org/mailman/listinfo/lsr

Reply via email to