[Isis-wg] draft-ietf-isis-traffic-01.txt

[Tony Li]

Don,

|  > There's no reason that you can't run the traffic engineering TLVs in
|  > parallel with the existing TLVs.
|  It is an either/or situation though for the new 24 bit metric and the old 6
|  bit metrics, correct? 

Not strictly.  You can run both.  If you do, you probably want to advertise
the same value in the 24 bit metric as you do in the default 6 bit metric.
You still get to append all of the TE subTLVs that we've specified, you
just lose out on the dynamic range of the metric.

|  The new connectionless metric does not have a specific traffic 
|  qualifier (such as delay, throughput etc) but I would assume its 
|  a combination of throughput and delay. Is this ever specified ?

It is the default metric, with identical semantics to the default 6 bit
metric, except for the range.  This is to say that it's an administrative
value, which may or may not have any bearing on reality, on a per-network
basis. 

|  We implemented thoughput and delay based routing on both a connectionless
|  and a path oriented system. 
|  The result for the connectionless system was that the delay based routing 
|  occupies forwarding table space for all possible destinations. 
|  We ended up implementing only two metrics. But with MPLS it is only 
|  a path minimization criteria during LSP selection.

Are you thinking of modeling the static delay or dynamic delay?  I'll
assume the latter.

One of the significant concerns with this type of approach is the stability
of the routing system when you impart dynamic metric information into it.
This has been discussed within the WG before.  Previous experiments from
the QoSR WG have demonstrated that we haven't been able to apply sufficient
control theory to the protocols to achieve stability.

|  > |  In my 
|  > |  experience multiple traffic engineering metrics for path 
|  > selection allow 
|  > |  better traffic engineering capabilities. 
|  > 
|  > Please say more.  The customer base didn't request this flexibility
|  We are defining TE beyond just ISIS. In general it depends on the size 
|  of the network and the diversity of the links. The bandwidth allocation
|  records are a type of metric in a sense since they take the place of the
|  typical throughput based metric in a connectionless system. Of course
|  not all MPLS connections will specify traffic parameters.
|  I would be quite willing to elaborate on what metrics I think are useful 
|  and why in a small draft for the next IETF.

Thanks, we'll take you up on that.  Also, if you'd like to lead a
discussion and argue your case with the entire WG, we can do that.

|  > One of the great glories of TLV encoding is the flexibility for future
|  > extensions.  ;-)
|  
|  Agreed, but can also be one of its curses if we don't do it right.

History has shown that we never get it perfectly right, so we do our best
and move on.  ;-)

Tony