cliche

A curious one this. I'd love it if someone understands and can help!

I've set up a wireless LAN (Belkin 54g router) round the accommodation of our hospital to provide a shared 1.5Mb cable connection for the residents. It's working fine. Except for one computer (Desktop, Win2000, USR 22Mbps 802.11 pci card), which consistently has slow (but working) access.

Computers in the same room (and further) can easily get full speed, but this computer seems limited to about 100kbit/s for its wifi connection. I briefly tried moving it next to the router (well, within a foot) but it made no difference.

Strangely, experimentation with ping -l 40000 (sending extremely large ping packets to test throughput) to the router shows that when the network connection first starts up, it runs at full speed. I can use outlook express on the computer to download emails etc from the internet, and the connection remains at full speed. However, when loading Internet Explorer approximately the first two or three connections (nb connections not pages/clicks) will load at full speed and then the wifi connection seems to 'switch down' to a remarkably constant 100kbits. This instantly results in a massive increase in ping times to the router as observed by leaving a command window running with ping -t -l 40000.

The speed will instantly return to full on disabling and re-enabling the pci device - but slows back down after the next few IE connections.

I'm not sure it's IE doing this, rather than something to do with bandwidth use - but I've managed to suck a fair bit of width (up to 500kbits) with massive pinging sessions without slowing the connection down. I'm going to have a look for other ways to cause the slowdown later - eg ftp or another web browser.

But, in the meantime, any ideas?

__________________
I can't understand why people are frightened of new ideas. I'm frightened of the old ones. -- John Cage (1912 - 1992)

Mephisto2

Sounds very strange.

Can you try opening several FTP sessions? See if the "performance hit" occurs with traffic other than HTTP. You may also want to try different browsers, just to see if that makes any difference.

I was going to warn you about WiFi and the medical industry; specifically your obligations under HIPAA, but I see by your profile that you're in the UK. I would investigate patient privacy regulations though if I were you.


Mr Mephisto

Cynthetiq

The Sarasota Hospital deployment of WiFi enabled PDA and laptops were pretty well protected upon simple look when I was there last summer and poked around on my PDA. They are one of the first hospitals to have gotten on board with electronic systems. I'm going to one day go to Memorial Sloan Kettering here in NYC as they are also part of the same program.

__________________
I don't care if you are black, white, purple, green, Chinese, Japanese, Korean, hippie, cop, bum, admin, user, English, Irish, French, Catholic, Protestant, Jewish, Buddhist, Muslim, indian, cowboy, tall, short, fat, skinny, emo, punk, mod, rocker, straight, gay, lesbian, jock, nerd, geek, Democrat, Republican, Libertarian, Independent, driver, pedestrian, or bicyclist, either you're an asshole or you're not.

Mephisto2

Yep. HIPAA requirements are quite strong and any medical organization in the US deploying wireless has to take this into account.

Mr Mephisto

cliche

It's for personal use - hence accommodation areas only.

I'll have a go at trying to find other things that can flip it into slow mode later. I'm feeling utterly confused by the whole thing, but there has to be a way to keep it at high speed bearing in mind it will quite happily bounce large ping packets around for hours until IE 'flips' it..

I had a quick play with tcpdump and I can't find any obvious differences before and after the flip - the tcp packets seem to be in roughly the same number of fragments, just slower.

EDIT

Update - it seems that any TCP traffic is capable of causing this slowdown. Accessing the router's own webpage can do it, ftp connections to internet sites can do it, and today a POP3 connection managed to do it. However, large pings to the internet (sorry, www.dslreports.com) won't do it despite using reasonably large amounts of bandwidth. However, once it's happened it affects ICMP the same as TCP.

I have a TCPdump log of it happening during an ftp session but there isn't anything obvious to see - packets remain the same size (1260) and windows don't alter. The only thing that changes is the times on the packets as they get further apart due to the reduced throughput.

Please, any ideas?

__________________
I can't understand why people are frightened of new ideas. I'm frightened of the old ones. -- John Cage (1912 - 1992)

Mephisto2

Update the firmware on the Belkin?

I'm grasping at straws here now.

Belkin are not exactly renowned as being top quality kit. :-/


Mr Mephisto