Hi Noel,

>> I think making sure that LISP is flexible enough to last a long time is
>> a huge benefit.
> 
> I basically agree with that, but we don't have infinite code-writing cycles
> available. (Remember, this is not an paper design exercise, this stuff is in
> significant experimental use.)
> 
> I can dream up dozens of things that would help make LISP "flexible enough to
> last a long time", but are we going to do them _all_ Right Now? Somehow I
> don't think so.
> 
> Resources for writing code are limited. We have to pick what is most
> important. The ability to support multiple dynamic EID blocks, without
> changing config files - something we're unlikely to use for a long time, if
> ever - is way less important than any number of more desirable upgrades.

I understand that. Auto-detection of those prefixes is a nice-to-have feature, 
but certainly not necessary for now. What if it's implemented as a configured 
prefix-list?

> That's really the bottom line here: saying 'we should do this' is equivalent
> to saying 'this is more important than X, Y and Z'.

Not hard-coding some prefix in code certainly is as far as I am concerned. 
Making extensions to the DDT protocol to auto-configure the prefix list can 
wait.

> (Add to that the fact, as far as I know of, only one organization is even
> putting energy into doing a commercial LISP implementation at the moment.)

Either you're talking about my company or you now know of two :-)
I am now running on old c7200's for a proof-of-concept, but those will be 
replaced by (probably) ASR1k's as soon as I start to offer it as a commercial 
service.

>> compare a few days/weeks of extra development work to years and years
>> of everybody being limited by a bad decision now.
> 
> It seems to me entirely unlikely that changing from i) a single hard-coded
> block to ii) multiple blocks allocated via some sort of dynamic mechanism is
> something that would be _particularly_ hard to add later - as opposed to,
> say, adding the ability to cache mappings in the MR (which requires changing
> the format of the Map-Reply message).

I am very scared of hard-coded stuff :-)  I would be very happy if it's not 
hard coded but just configurable in some way! I agree that changing existing 
message formats later would be so much worse.

> Getting the correct balance between i) getting stuff up and running, and ii)
> not handcuffing yourself in the long run, is a tricky problem. I lived through
> one cycle of this (with the initial IPv4 stuff), and lived to see _just how
> painful_ what one thinks is a 'quick hack to get us up and running' can be
> (with the 32-bit addresses).
> 
> I am never going to make that mistake again.

:-)  Happy to hear that

> This is not in that class.

We might disagree a bit on that, but let's look for workable solution. From an 
operational point of view I really don't like hard-coded address blocks that 
have a special meaning, but I am also certainly not suggesting that you drop 
all other work! Would using a prefix-list to configure this be a good solution 
for now? (I suspect there must be existing code to check an address against a 
prefix-list somewhere ;-)

Thanks!
Sander

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

Reply via email to