Charlie
Stan, Vicky and I (not forgetting to mention Rick) have spent five years trying
to work out how to get bandwidth and latency metrics from modems into routing
protocols, so those metrics can be used to direct traffic in some fashion.
There are lots of opinions on how best to do the directing so I’m not even
going to go there, but suffice to say that I believe we should make an effort
to incorporate outputs of DLEP into AODVv2. I will be the first to agree this
is a non-trivial task, so I would be happy to vote for making this feature a
separate draft.
On the other hand, from some observations of colleagues work in this area, I
would suggest that a collection of metrics are often used to direct traffic
according to some mathematical function. I would suggest it is not sensible to
use bandwidth alone to prefer one route over another in the way that hop count
has been used for decades, but most effectively when factored into such a
calculation. Perhaps we should add some text on this point, which could then be
used to placate the critics.
Regards
John
On 28 Sep 2015, at 06:38, Charlie Perkins <charles.perkins@xxxxxxxxxxxxx>
wrote:
Stan,
It's not a problem at all to construct routes where hop count is not the best
metric.
Was this at issue? Where did I say hop count was always best?
Or were you trying to suggest that bandwidth, as a non-cost metric, was
sometimes useful? Well, yes, of course it is.
There are well-known cost metrics that enable comparisons of routes to select
the one with the best bandwidth. I have mentioned them before in
conversations on this list about this subject.
If my part of the discussion is simply to be dismissed as protestations, I
wish someone would just tell me. I'd much prefer to know how to participate
in a constructive discussion.
Regards,
Charlie P.
On 9/27/2015 2:52 PM, Stan Ratliff wrote:
On Sun, Sep 27, 2015 at 5:31 PM, John Dowdell <john.dowdell486@xxxxxxxxx
<mailto:john.dowdell486@xxxxxxxxx>> wrote:
Charlie
On 24 Sep 2015, at 21:48, Charlie Perkins <
<mailto:charles.perkins@xxxxxxxxxxxxx>charles.perkins@xxxxxxxxxxxxx
<mailto:charles.perkins@xxxxxxxxxxxxx>> wrote:
Hello John,
Follow-up below...
Bandwidth is not a cost metric. The calculation of LoopFree() might be
more complicated. I don't like non-cost metrics for various reasons. One
reason is that they contradict the entire mathematical formulation for
metrics and metric spaces. On the other hand, some people in the IETF
don't have much respect for mathematics, it seems, and I get tired of
complaining about it.
Regards,
Charlie P.
Bandwidth is a metric (along with latency, both of which are declared as
metrics in DLEP). You can use it to influence routing decisions, but I’ll
agree that it doesn’t work to completely drive a route selection in the way
that hop count does.
So, let me construct a network scenario where I have two paths to a device:
Source------Router1--------Router2---------Router3-------Router4-------Destination
But, all of the dashed lines in the above description depict 10Gb/sec
copper.
Second path:
Source-----Router1-------------------Router3-------Router4------Destination
But the long dashed line between Router1 and Router3 is a satellite backup
link, at 1Mb/sec. I understand - it's a hypothetical case. But stuff like
this pops up all the time in real networks.
First route, hop count 5. Second route, hop count 4. Based on Charlie's
protestations, I should assume the second route is better than the first?
FWIW, I don't. If there's any way on earth I can take the first route, I
will.
Stan
I believe it is also impossible to calculate LoopFree() just using
bandwidth, so maybe you’d have to use a combination of hop count and
bandwidth, but that is a discussion for another draft.
Regards
John
