[Preston] Network Newbie need help with WiFi

[Ken Wolstencroft]

Hi Richard,

I am trying to avoid an ad-hoc cloud, it likely to be unreliable.

I prefer the idea of an ESS, would I be correct that all Access Points need
to have the same ESSID in order to operate in this mode (e.g. 20 access
point all with the same ESSID of wnap).

Thanks,
Ken
WNAP

----- Original Message -----
From: "Richard Revis" <richard@revis.org>
To: <preston@mailman.lug.org.uk>
Sent: Wednesday, November 20, 2002 10:52 PM
Subject: Re: [Preston] Network Newbie need help with WiFi


On Wednesday 20 Nov 2002 9:52 pm, Ken Wolstencroft wrote:

> St Annes is not a problem, the idea is to build a number of nodes and then
> fill in the gaps.

Well, I *might* be able to nick a 24dbi directional antenna from work, to
point at a tower block in Preston ;o)

> I was considering using zebra, but I will have a look at mobile mesh.

Zebra/OSPF has a number of issues, the worst being the application of routes
to the kernel routing table on backup routers. OSPF is also not designed for
a wireless environment and as such the metrics are not the best. You may
have
seen my posts to $mailing_lists about my issues with Zebra recently.

On the plus side you do not need to run the routing software on every node -
I
have a perl script which uses a broadcast ping to detect clients, which then
adds them to the kernel routing table and this is distributed via OSPF.

Grid (http://www.pdos.lcs.mit.edu/grid/) *does* take account of link
congestion as well as hop count but is even worse to configure than Zebra
IME, and is not ideal for large scale WLANs, especially since it seems very
prone to partitioning for no reason I can determine.

Mobile mesh (http://www.mitre.org/tech_transfer/mobilemesh/) is a happy
medium. You *do* need to run it on every node, but it is very very easy to
run (even on a zaurus). It can be configured with multiple internet
connection points accross the WLAN if required, and can use these to route
WLAN traffic via wired routes if it determines that this would be faster.
This also allows it to mostly stop BSSID partitioning, a serious problem in
very large WLANs. The metric system is not the best but in IME works fine in
99% of cases.

Where hotspots are not linked of course you can then route your
authentication
information over the global internet to the radius server(s).

...this all assumes that you were planning to run the thing as a large
ad-hoc
cloud of course.

> At present I have been building a central user database for radius
> authentication. This will allow users to access any of our groups nodes
> from a single user account, the idea is to connect users via vpn.

IPSEC or CIPE? Be wary of tcp over SSH where link quality is bad -
http://sites.inka.de/bigred/devel/tcp-tcp.html

> I would like to build a list of potential nodes for the group, so if
anyone
> is interested either e-mail me direct or to the list. The group website
> will be launched at the end of the week, so if you would like to join the
> fun I will add you to the list of nodes.

As you may have gathered I did do some of this for profit as well as fun ;o)
Hope some of this information helps you, after all I got paid to bang my
head
against a brick wall, no reason why you should have to do it again for free!

--
Richard Revis
Outgoing e-mail is signed with my public PGP key
You may obtain a copy of my key at http://key.revis.org

_______________________________________________
Preston mailing list
Preston@mailman.lug.org.uk
http://mailman.lug.org.uk/mailman/listinfo/preston