RE: Next question...

["Tony Li" <Tony.Li@procket.com>]

Tony,


|   > My concern is not so much about loss of control as it is with 
|   > the administrative burden of implementing the policies.  
|   
|   Isn't perception of burden really a function of direct vs. indirect
|   control? 


Umm... no, it's a function of the number of unique operations necessary
to implement and maintain the desired policies.

   
|   > Most policies are going to be site-wide, not host specific.  
|   > Having to distribute this policy and keep it consistent 
|   > across an enterprise can be done by our standard system 
|   > administration tools, but the effort involved in doing so is 
|   > non-trivial and the number of exception hosts is likely to be 
|   > small, so this just seems like a poor ROI for the site 
|   > administrator(s).
|   
|   I agree that most policies are site wide, but ROI depends on how
|   valuable the exception list is. If you are in the exception 
|   list, your
|   perspective is different than if you are not. As far as 
|   triviality, that
|   depends on the mechansim. If it were something like a 
|   static policy on
|   all host to prefer the prefix order in the RA, it seems 
|   pretty trivial
|   to me.


Such a policy is trivial regardless of the location of policy implementation.
The question is about the effort to distribute that policy.  If it's
on the SBR's then it's less than ten systems.  If it's on the hosts,
then it's thousands and any rogue sysadmin can abuse it.
   

|   > Our job, as architects, is not to create an architecture with 
|   > infinite flexibility.  Rather, it is to reject those options 
|   > which lead to unnecessary complications down the line.  
|   
|   And our disagreement really comes down to definition of 
|   'necessary'. 


True.  Why is individual host control 'necessary'?

   
|   > Control at the host seems like an infrequently used feature.  
|   
|   Frequency is not an indication of value. Besides it is done on every
|   connection today, so that seems pretty frequent to me. 


AFAIK, no host has the wherewithal to make the intelligent
routing decisions that you're proposing.  Which connections are you
thinking of?

   
|   > Note that policy for a specific host can be implemented 
|   at either the 
|   > site border router (SBR) or at the host, so removing the 
|   > policy decision from the host isn't actually eliminating 
|   > architectural flexibility.
|   
|   I absolutely disagree. Hosts actually make those policy 
|   decisions today,
|   and the system works. The machine I am typing this on has 4 IPv4
|   addresses, yet it has no trouble opening a connection to CNN.com by
|   picking from that list of possible sources, and the 8 available IPv4
|   destination addresses. There seems to be a fear that providing more
|   choices to the host will magically make the world more complex. 


No.  Providing the host with the information so that it can make the
choice about 8 destination addresses and 4 source addresses is part
of what's scary.  That and distributing the policies.

   
Tony