[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [RRG] Routers in DFZ - reliable figures from iPlane

Hi Ricardo,

Thanks for your response.  I asked Harsha to join the RRG list and
respond to you, since I don't know enough about this to answer your

You wrote:

> a stub can be multihomed and only use half of the routing
> table, the other half use default routing.

"Multihomed" to me means it has two or more links to "the rest of
the Internet" meaning that it has to choose between these two or
more links for packets addressed to every prefix except those of the
local network(s) its other links lead to.

"Stub" to me means it has one link - so I don't see how a router
could be a stub and multihomed.

> What you call singlehome may not be singlehome because of the limited
> visibility of iPlane.

Hartha would need to comment on whatever limits there are to the
reach of iPlane's research.

> 2. These numbers are at most a lower bound on the numbers you are
> looking for, and we have no way to tell how tight this bound is. iPlane
> data is incomplete because of several reasons, such as limited number of
> vantage points, blocking of ICMP/UDP by routers, MPLS tunnels,....

I can imagine there must be some limitations on what any research
project could achieve.

> Furthermore, there's the problem of converting router interfaces to AS
> numbers which is not entirely resolved, see these papers:
> http://irl.cs.ucla.edu/topology/chang01inferring.ps
> http://irl.cs.ucla.edu/topology/fwdPath.pdf
> http://irl.cs.ucla.edu/topology/3rdparty.pdf

I can't comment on how iPlane tackles these problems.

>> We are currently considering a major, kludgey, mapping and tunneling
>> overlay architecture to add to the Internet, in addition to moderate
>> and generally improvements to the BGP protocol itself.  Before
>> anyone could do a cost-benefit analysis of this, they would need an
>> estimate of how many DFZ routers are afflicted by the problems which
>> prompted us to propose such a drastic solution.  They would need to
>> roughly characterise those routers in terms of their model number,
>> number and capacity of interfaces etc. - and also their cost.

> And getting realistic numbers for those is an unfeasible task as of now,
> at least for the entire Internet. My advice is to get accurate data for
> a tier-1 ISP directly and start from there.

How would knowing the number of routers for one large ISP help
determine the total number in the Net?

>> According to my understanding of this file, each line represents a
>> BGP router in the reasonably complete set of traceroutes the iPlane
>> project works from.  Each line contains an IP address for each of
>> its interfaces.  So I thought it would be easy to count these, to
>> find how many had a single IP address - those would be single-homed
>> routers and the rest would be "multihomed border routers" and
>> "transit routers" in the DFZ - all with two or more BGP peers.

> Nope. A router can have multiple interfaces connecting to same provider.

Sure, but if they go to different routers, then the router needs to
do the full BGP operation and compare notes with these two or more
peers, decide on the best path for each of 220,000+ prefixes etc.
As far as I know, it doesn't matter which provider the router's
peers are in - but please correct any mistake I am making, since you
and most other people on this list know more about this than I do.

  - Robin

to unsubscribe send a message to rrg-request@psg.com with the
word 'unsubscribe' in a single line as the message text body.
archive: <http://psg.com/lists/rrg/> & ftp://psg.com/pub/lists/rrg