On Tue, 2 Aug 2005 12:23:11 +0200, "Doug Johnson" <doug@johnson.net>
wrote:

>Sorry for using the wrong expression. Say I have a teoretical WL link
>bandwidth of 54 Mbps (maybe more because of the GS version) and expect 8
>users to be connected simultaneously using WPA encryption. In my original
>question I wondered how much of those 54 Mbps will be eaten up by
>handshaking, encryption, timeslicing etc,etc.

Yeah, that's better. 54Mbits/sec is the connection speed. Since it's
a simplex (tx and rx one at a time) system, you'll never get
"wirespeed" performance. In addition, all the 802.11b/g management
overhead and necessary inter-packet delays add substantial loss in
thruput. I once did a chart of the limits.... digging....

This is stolen from an Atheros PDF at:
http://www.atheros.com/pt/atheros_range_whitepaper.pdf
with some additions and corrections by me.

Non-overlapping Modulation Max Max Max
Channels ------- | Link TCP UDP
| | | | |
802.11b 3 CCK 11 5.9 7.1
802.11g (with
802.11b) 3 OFDM/CCK 54 14.4 19.5
802.11g only 3 OFDM 54 24.4 30.5
802.11g turbo 1 OFDM 108 42.9 54.8
802.11a 13 OFDM 54 24.4 30.5
802.11a turbo 6 OFDM 108 42.9 54.8

The paper claims that encryption is enabled for these calculations,
but my numbers seem to indicate that these number are for encryption
disabled. Dunno for sure. The Max TCP and Max UDP are the
theoretical maximum thruput rates. If you want the detailed calcs, I
think I can find my notes. I don't have numbers for the various Turbo
and Super modes.

WPA encryption eat about 10-15% additional loss in thruput.

If your backhaul (DSL or Cable Modem) speed is less than the maximum
thruput speed, the backhaul will be the limiting factor in thruput.

All the above is for ONE user. There's no additional overhead for
adding additional users. For an equal speed wireless connection, the
thruput will simply divide equally among the 8 users. It gets messy
if each user connects at a different speed. The algorithm is in my
never humble opinion, rather stupid. The access point distributes the
bandwidth by equalizing the number of packets. A 1Mbit/sec slothish
connection will get 1 packet, while a 54Mbit/sec will also get 1
packet. However, it takes the 1 Mbit/sec connection 60 times as long
to deliver the one packet as the 54Mbit/sec connection. That means
the bandwidth of the 54Mbit/sec connection is drastically reduced in
the presence of a 1Mbit/sec connection. Not all access points work on
this principle, but the cheap ones tend to do it this way because its
easier to implement. The "fair share" algorithm is run at ethernet
speeds in the router section, and the wireless part just pumps out
packets.

>You mention that one bittorrent user will choke all the others, does this
>mean that the wless protocol does not have any load-balancing between
>clients other than setting priority by IP or MAC address from the WRT's
>config utility?

One BitTorrent user will not exclude everyone else. The problem is
that BitTorrent will monopolize ALL the available bandwidth. Usually,
it's the outgoing bandwidth that gets filled first. You could have
all the incoming bandwidth you want, but if the TCP ACK's are being
constipated or delayed by filling the outgoing bandwidth, incoming
performance will truly suck. At one ISP, we fired off a BitTorrent
download of a Linux CDROM image and successfully slowed a fractional
OC-3 (about 45Mbits/sec) backhaul to the point where alarms were going
off. BitTorrent opened enough simultaneous streams to perform the
download very quickly, but did so by literally monopolizing the pipe.

The solution is to use some form of bandwidth management or QoS.
Limiting the available bandwidth by user or service is in my opinion
mandatory. I also reserve some bandwidth for VoIP which is very time
sensitive. I have one system using Snort to detect hacking and abuse,
but also to detect use patterns that tend to monopolize the bandwidth.
Due to lack of time (and interest) I haven't done much with it.

>Even if there are network transport request from other
>users during one client's file transfer the others will not be granted
>bandwidth before the transfer is terminated?

Huh? I don't understand the question. Without QoS or Bandwidth
Management, wireless bandwidth is mostly allocated on a per packet
basis in rotation. It's a really brain dead way of doing it. There's
no intelligence on the IP layer or above because wireless is 100%
bridging, which is all at the MAC layer. At the MAC layer, there's no
concept of users, sessions, files, ports, services, etc. Just
packets. That's why QoS and BW has to be done at the IP level to be
effective. I'm not sure I answered your question, but if you rephrase
it, I'll try again.

>Do different wl routers / access points handle this differently that the
>WRT54? I'm not thinking of expensive Cisco eqpmnt but boxes in the sub $500
>'area'.

I have no idea. Wireless routers have three sections:
- A wireless access point (bridge) radio:
- A router that handles the LAN - WAN stuff:
- An ethernet switch to allow local connections:
The wireless part is mostly a tangle of encryption and timing. Take
away these complications and all wireless bridges are fundamentally
the same. There are some differences in implementation at the MAC
layer, but the timing and protocol rules are set by FCC part 15 and
IEEE 802.11b/g, and do not really vary much by manufactory. There is
considerable differences in radio quality and chipset features, but
that's not part of this discussion.

The ethernet switch is fairly simple except that some wireless routers
(WRT54G) allows independent router control over each switched port.

It's the route features that distinguish the cheapo wireless routers
from their overpriced versions. In general, the fancy routers offer
much more control over IP layer packets than the cheapo routers.
However, you can get the same (or similar) features by simply buying
an access point (or buying a cheapo wireless router and configuring it
as an access point) and adding a decent Cisco ethernet router. I'm
fairly sure all of the important wireless router features are also in
the ethernet routers.

What you apparently need/want/interested is bandwidth management and
that's a router feature. There's nothing in the bridge/radio section
that does anything useful for bandwidth management. Note that Cisco
has mostly fairly simple access points and relies on their routers to
do the real work.

>The scenario is one office network connected to a WRT54GS in AP/bridge mode.
>6 users will each have a WRT54GS (in client mode) connected to their
>computer at home. They will do casual business tasks and internet access.
>Filesharing and other high-demand bandwidth ports are closed at the office's
>firewall. Would the linksys boxes do sufficiently in this case? If
>marginal, do you have other suggestions?

Sorry. I don't understand your description. I also don't understand
how this question relates to your first question and cannot tell if
you're building a new network or trying to solve a specific problem on
an existing network. How does the home user connect to the office
network? Is there a VPN involved? If so where does the VPN
terminate? What are the bandwidths available to the home users
(dialup, ISDN, DSL, etc)? Are you seriously worried about abuse by
your own clients or is this a performance issue? Please organize your
questions in terms of "What problem are you trying to solve or what
are you trying to accomplish" and then supply "What do you have to
work with?".


--
Jeff Liebermann jeffl@comix.santa-cruz.ca.us
150 Felker St #D http://www.LearnByDestroying.com
Santa Cruz CA 95060 http://802.11junk.com
AE6KS 831-336-2558