[arin-ppml] "Millions of Internet Addresses Are Lying Idle" (slashdot)

David Farmer farmer at umn.edu
Mon Oct 20 19:20:41 EDT 2008

On 20 Oct 2008 James Hess wrote:

> On Sat, Oct 18, 2008 at 4:22 PM, Leo Bicknell <bicknell at ufp.org> wrote:
> [...]
> > However, the larger problem is that this is the entirely wrong
> > definition of "in use".  Let's take a simple example of a University.
> > They may have wired dorm rooms, students with laptops, and wifi
> > enabled classrooms.  If you ping at 11AM, the classroom IP's respond,
> > and the dorm ones do not, if you ping at 11PM, the dorm rooms
> > respond, the classrooms do not.  And if you ping at 12 noon when
> > they are all walking to lunch almost none of either respond.
> [...]
> It is a flawwed methodology to use ICMP / TCP pings  to detect hosts

I agree this is a flawed methodology.

> The inefficiency results from tying up the classroom WiFi   ips at
> times when these IPs are not needed at all  and tying up the dorm IPs
> at times when they are not needed.
> This inefficiency may result in consuming twice as many IPs as needed;
> since each host essentially has a different IP at different times of
> the day. 

Remember that most laptops have at lest two MAC addresses, one for the 
Wireless port and one for the Ethernet port.  There is no easy way to 
correlate the two.  So at least some of the inefficiency you refer to, is built 
into the architecture of most PCs, and beyond the control of a campus 
network operator.

> Efficient use would be to have one pool of IPs;  a laptop is assigned
> an IP from a DHCP pool, That pool is the same for both classroom WiFi 
> and for dorm room connectivity. And does not change if a laptop is
> moved between two places on the campus net. 

A dorm network and a classroom network, Wifi or not, are two different 
networks they have two different purposes.  One is a residential network the 
other is essentially a corporate network.  A campus network operator plays 
much different roles in the two networks.  In one case we act much like a 
corporate network, in the other we are much more like a residential service 

If Comcast provideds me business Internet at my work place and 
Residential Internet for my house, would you expect them to conserve IP 
addresses and provide me the same address in both places?  Some how I 
don't think so.  So why are you expecting this from a campus network 

> Logically, the IP is a property of the host. The reason a host would
> ever have different IPs  when plugged into different parts of the same
> organization's  network is that the topology is laden with an excessive
> number of Layer 3  routing devices. 

Maybe yes, maybe no. 
> Instead,  switches  and bridges  should be used to connect the   Dorm
> and classroom WiFi networks. Any   firewalling,  broadcast filtering, 
> traffic limits, DHCP use enforcement,  etc, between the two  should  all
> be implemented on  transparent bridges. 
> This  design would eliminate unnecessary duplication of IPs for the
> small set of hosts.

I don't think you understand the scale of many University networks, for 
instance the network I'm responsible for covers 1,233 acres, almost 250 on-
net buildings, containing 21.2M gross sqft, with over 80,000 GigE access 
ports, servicing over 50,000 students and over 17,000 staff and faculty. 

I realize we are one of the larger university campus networks, but there are 
probably about 50 to 100 university networks that are at least half our size or 
greater, and that is still a fairly substantial network.  
So when you are talking about this scale of networks some amount of Layer 
3 topology isn't just an option it is a necessity, most enterprise class closet 
switches have a 16K or 32k CAM table size.  We have a Layer 3 backbone 
by design, to allow us to scale to this size.  We design for 2K to 5K access 
ports within a Layer 2 switching domain, this leaves room for running over a 
bit we have a few domains with more like 7K access ports and for multiple 
MAC addresses on an access port, like for Wifi-APs, VoIP phones with PC 
ports, and those small desktop switches that people install without telling the 
IT or networking staff. 

So just like commercial Internet operators, one size doesn't fit all and 
making to many assumptions can get you in trouble quick.

David Farmer				     Email:	farmer at umn.edu
Office of Information Technology
Networking & Telecomunication Services
University of Minnesota			     Phone:	612-626-0815
2218 University Ave SE			     Cell:		612-812-9952
Minneapolis, MN 55414-3029		     FAX:	612-626-1818

More information about the ARIN-PPML mailing list