Jim Schaad wrote:
> I have some comments on the draft.  However first let me say that I know next 
> to nothing about RADIX having only briefly read it a couple of times and 
> having not yet worked with it to any extent.

  Reviews are always welcome.

> 1.  I think a discussion is in order about the issues of packing a radius 
> message full in the presence of proxy intermediaries which might need room 
> for adding and removing routing materials.  I don't know how many proxies are 
> keep local state as oppose to just filling up the message with Proxy-State 
> items and thus need space on the way back.  Also a discussion of removal and 
> re-insertion of the T flag might be needed if the proxy fully assembles the 
> message and then relays it in pieces.

  The simplest way to address that is to artificially lower the RADIUS
maximum packet size, as seen by the client.

> 2.   The Overview in section 2 does not discuss the fact that a new set of 
> state items might also added to the message.  This changes the limit of how 
> many bytes can be placed into the fragmented packets.

  Yes.

> 3. nit - in para #3 of overview you use the word truncated and truncation.  I 
> think a better word set might be partitioned and partitioning.  Also ensure 
> that fragmentation is used strictly for the process in radius-extensions.  I 
> have not found any cases yet were it is not but will try and remember to look.
> 
> 4.  I am not sure if positioning of the more-data-pending packet is 
> significant.  The introduction says it comes last but nothing to that effect 
> is said in section 3.  
> 
> 5.   Section 6 implies that the order of items in a set of fragmented packets 
> might need to be changed based on the ability of items to be in the packet.  
> For example if the Access-Accept packet had been = Data1[M], Data2[M]...Data 
> 10, User-Name, Service-Type[X], Framed-IP-Address the order would need to be 
> changed in the original partitioning.  This should be mentioned someplace 
> other than just in the "examples" section.  It implies that a general use 
> piece of code to do this will need to have the table built in, and 
> potentially extended as new items are added, rather than just being coded 
> correctly in the specific packet type code.

  Attribute ordering is explicitly *not* required in RADIUS.

> 6.  What if any indications does a server have that the client will be able 
> to deal with the partitioned message other than it will either fail or get 
> confused depending on what set of attributes are in the first packet and what 
> it needs and feels it can ignore.

  That's an open area for discussion.  RADIUS doesn't have capability
negotiation, so it's hard to do.

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

Reply via email to