Dave Hartzell wrote:
> I think we would be *far* away from this seamless switching without
> Mobile IP. The problem today is that most wireless network support
> portability, not mobility. Meaning you need to reboot or get a new DHCP
> address when you roam between APs connected to different Layer 3 networks.
Mobile IP is not necessary (or even very desirable). 802.11 supports
mobility at layer 2. You just need smart enough access points. It is
possible to produce smart enough APs in Linux or {Free,Net,Open}BSD,
using the HostAP mode. I will explain.
I don't think Mobile IP is a very compelling mobility solution if your
mobile PC is only a client, never a server, and if you can tolerate
no level-2/3 security. I think that this describes most applications a
mobile PC will see, since mobile's bandwidth will not usually suit serving
anything, and since the secure socket layer will serve most of a mobile
unit's security needs. Also, it seems complicated to deploy/administrate
the infrastructure for Mobile IP, and to install/support the client
software---is it not?
The 802.11 spec already provides for a mobility solution and, in fact,
all the software an 802.11 client needs for mobility is in the firmware
for its 802.11 client adapter. All the "work" of supporting mobility,
like coordinating "hand offs" from cell to cell, and forwarding traffic,
an access point can (and should) do. You could roam with your laptop
tomorrow, *without installing any new software on your laptop*, if there
were enough allied access points *with enough intelligence* to make sure
that your traffic followed you.
The way that this would work is this. 802.11 Associate/Reassociate
messages, which are in the vocabulary of every 802.11 client adapter, are
sent from a client to an AP when a client wants to select or re-select
its connection to the wired/wireless infrastructure and thence to the
Internet. Whenever a client receives a stronger signal from a different
access point, AP B, than its present network gateway, AP A, it sends a
Reassociate message to AP B. In the Reassociate message is the identity
of AP A. At this point, intelligent access points would "hand off" the
mobile station, arranging for traffic bound to the mobile station at A to
be sent instead to B. They would do this even if the APs were on different
IP subnets. APs are missing three features that would let them do this:
1 an Inter-Access Point Protocol (IAPP) which will let them coordinate
hand-offs over IP networks
2 the smarts to use IP as the 802.11 Distribution System (DS)
3 a Rendezvous Service (RS) with which APs in the same ESS discover
each other for the purposes of talking by IAPP and addressing each
other on the DS.
I don't think there is any such access point, yet. Some people do not
realize that no particular Distribution System implementation is built
into the 802.11 spec. (What about WDS, you ask? WDS a distribution system
does not make!). The 802.11 spec is deliberately vague about the make-up
of a DS. The DS could be a switched Ethernet LAN, a multi-hop wireless
network, or an IP network. Likewise, 802.11 punts on the Inter-Access
Point Protocol. An IEEE "draft recommendation" describes a (sort of
abstract) IAPP, suggesting UDP and Ethernet implementations. But I have
not discovered any vendor that uses an IP distribution system, and while
I believe some vendors (Cisco? Lucent?) implement an IAPP that operates
over Ethernet only, I doubt that any two vendors' IAPPs will interoperate
on Ethernet, let alone IP. I am not aware that any standards drafts
describe a Rendezvous Service for access points, even in the abstract.
I know there is the talent among enthusiasts of community wireless
networking to produce the access point software (using the Prism HostAP
mode, say) for IP mobility in open source. I am excited to work on
this myself, but speaking realisticaly, I will not get around to it for
a while.
Dave
--
David Young OJC Technologies
[EMAIL PROTECTED] Engineering from the Right Brain
Urbana, IL * (217) 278-3933
--
general wireless list, a bawug thing <http://www.bawug.org/>
[un]subscribe: http://lists.bawug.org/mailman/listinfo/wireless