Tcp/Ip wireless throughput

[Kernalx69]

Hi

I have a Cisco pix 515e firewall and a Cisco Aironet 1242AG
in a network with several computers and servers.
I have made a vlan that the wirelss traffic should go through.
For wirelss users to connect to the wired network they must go through the firewall.
All traffic i allowed for now.
Everything works fine except that the throughput is very bad in some cases.

If i take a wireless computer and download a big file (100MB) from a wired server 2000 and i measure the speed , the speed shows about 10 Mbit/s.
But if i make the same test to a Windows 2000 pro on the wired network it shows 22 Mbit/s.
It doesn't matter where i put the Accesspoint in the network the result is the same.

If i insted make the same test with a switch that i configured to the Vlan3 and connected a computer to it.
The measurement to the server shows 70 Mbit/s , and to the windows 2000 pro 60 Mbit/s.
As you can see in this case it is faster to the server.

If i don't use vlan the speed is ok.

Why is it like this?
What should i do ?
I wounder could it have something to do with MTU size or Tcpwindow size ?

Please help !

Best regards

 

[pansophic]

Are you asking why the connectivity is slower when you are in a different VLAN? And even slower if you go through the firewall?

If so, it has to do with the amount of processing that must be done on the Layer 3 devices in the network. The switch has to do some level of processing to route packets between the two VLANs, which adds milliseconds of delay to each packet. In the case of the firewall, packets must be processed at Layer 3 to determine if the connection is allowed, and compared against a state table. All of this processing comes at a price, and that price is speed. The firewall will add a few 10s of milliseconds to the packet processing.

MTU size and TCP window size can affect some of these processes in a positive way, but generally, the larger the MTU size, the fewer the packets, therefore a reduced delay in the overall transmission. But if packets have to be fragmented to get across another piece of the network, like your wireless LAN, then that savings can quickly be eaten up in fragmenting and reassembling packets.

Perhaps someone here has some concrete experience with good MTU sizes and TCP window sizes for this Cisco equipment?


pansophic

 

[Kernalx69]

Thanks for your answer.

I can understand that the throughput will suffer then i go through a firewall.

But i can not understand why the throughput is better to a workstation than a server then i go through a wireless vlan.
If the throughput on the server was god as on the workstation i had accepted that.
But the starnge thing too is that if i take a new server and make the same test , the result is the same.

I think that the default MTU sice is different between then the server and the workstation and then i use wireless the throughput will suffer.

 

[pansophic]

I doubt that the MTU size is different between Sever and Pro, but I guess that is a possibility. It is more likely that resource allocation is different between the two versions of the OS, and that as you increase the number of simultaneous sessions you would see the Pro performance drop off dramatically, but not with the server version.

Have you tried running 10 simultaneous connections to the test machine from 10 different workstations?

That is just a guess, but generally the performance of a multi-cpu or multi-user operating system is only seen when scaled, not on a 1:1 performance test.


pansophic