[MLB-WIRELESS] Re: [mesh] Transfer Rates

[Cameron Donaghey]

Hello All,
   Ok got stuck into it and did some various testing last night. It seems 
that I did overlook that
fact that transferring client->AP->client halves the bandwidth because the 
network is shared.
The maximum I can achieve with client->AP->client is 282kB/s which allowed 
me to transfer
a 20meg file in 1min11secs. This from what I now understand is quite good.

I then decided to take the access point out of line and try the two clients 
peer to peer ( adhoc )
or client->client mode and see what they could achieve by themselves.
The maximum I could get with the same 20meg file was 569.05kB/s which 
allowed me to
transfer the file in 35.5secs.

I then also tried running one client to the access point connected to the 
second machine
via its ethernet connection so client->AP->wired.
The maximum I achieved from this was 546.6kB/s and I transferred the 20meg 
file in 37.9sec.

So from all of this I think my setup is running how it should be although I 
still don't entirely
understand when going client->AP->client the transfer speed is halved when 
compared to
the client->client, perhaps someone could explain or perhaps this is not 
correct.

Thanks for everyone's replies,
    Cameron

At 10:35 AM 9/17/2002 +1000, Geoff Smith wrote:
>I may have mis-read the original post, I thought they were
>attempting client->client transfers not client->AP->client transfers.
>My apologies if my post mislead anybody.
>
>Here is another link:
>http://support.countryday.net/Wireless/Performance/Wireless%20Efficiency.htm
>it shows the lucent wavelan card getting 4.93Mbps, (616kB/s), for
>client->AP->wired  transfers
>
>client->AP->client transfers with an AP with a single channel radio will be
>sharing
>the same bandwidth between the two clients (i.e. half again) but more
>importantly
>its very sensitive to the capabilities of the AP, embedded os, cpu speed, 
>memory
>size
>for queueing and buffering etc etc; i.e. implementation issues. In 
>addition the
>people who
>designed the AP probably tuned it for client->AP->wired connections
>not client->AP->client connections.
>
>If the orginal post meant kB, then he was getting 2.4Mbps which is pretty
>impressive
>and would represent 21% of the total bandwith
>If the original post meant kbits then he was only getting 2.6% of the total
>bandwidth
>for a client->AP->client test compared to 44.8% achieved by the lucent 
>card in a
>
>client->AP->wire test, or 27% achieved by the wl100 card in client->AP->wire
>tests.
>
>So if Cameron is doing client->AP->client, perhaps he can tell us what
>he is getting for a client->AP->wire test?
>
>Regards
>Geoff
>
>
>
>Craig Mead wrote:
>
> > | Do you mean kB, k bytes or kb, k bits
> > | 310kbytes -> 2.4Mbps
> > |
> > | Theoretically you should be able to get about half of the 11Mb
> > | bandwidth.
> > | ~ 5.5Mbps or ~680 kbytes/s
> >
> > Doing client to client transfers, this is extremely innacurate. The 11b
> > protocol has some massive overheads which consume a high percentage of the
> > possible thruput. The highest I've ever seen a link running personally is
> > 570k/sec (client to wired server behind AP). Though theoretically from this
> > you should then be able to get ~ 285k/sec client to client thru AP, 
> which is
> > a lot higher than the 50KB/sec max I've obtained.
> >
> > I had a talk to some of the boys in Melbourne about this issue and they 
> said
> > it also depends on the placement of the end nodes in relation to each other
> > and the AP for some reason. Not sure about the technicalities of how this
> > would effect it, but they said it did. As soon as I get a few more links on
> > I'll be able to do some more testing.



-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.wireless.org.au/pipermail/melbwireless/attachments/20020917/131a7c8f/attachment.html>