Duncan Sands ([EMAIL PROTECTED]) wrote: > > Second question (see [1]): > > In my logs the hotplug even manager loads the speedtch driver trice. I > > don't know why. The problem is it fires up 3 modem_run. Thanks to > > modem_runs design, only one instance finishes the initialization. Isn't > > there a better solution ? > > This is due to a usb change in the kernel: it now (since 2.4.2x, x = something) it > launches hotplug for every interface, not each device. This is in line with 2.5 > behaviour. However, unlike for 2.5, the kernel does not export info about which > interface hotplug is being called for. I guess it is possible to work around this > using > lock files storing pids and all that jazz. One day I will send in a kernel patch to > export the extra info, it is on my infinite TODO list... In practice it doesn't > matter.
That's what i guessed. Thanks for the complete answer. > > Third question: I'm not sure the pppd launch belongs to the hotplug > > event manager. Hotplug is run before any network services, so the > > connection would be done before any firewall would be up... As you > > see it's not really a question, just a loud thought ;-) > > It is convenient though. You could launch pppd in an init script of > course and it wouldn't make much difference, except if you replug the > modem. By the way, won't your firewall barf if ppp0 does not exist > when you try to do (example) masquerading with it? No, I've never seen a firewall complaining about rules applied on unknown interfaces. It would be a clear implementation flaw if it did so, as it's good(tm) to prepare rules before the interface is up :-) Btw if you include a ppp interface upping stage in this script, you have to include a ppp interface downing stage, otherwise the ISP is just not aware of your disconnection through the PPP communication. This was a cause of problems in early aDSL days where IPs were allocated and never freed in the ISP side. Nowadays this problem is no more afaik. > > Fourth question: I noticed very CPU usage during network load > > (around 5%[2]) for a 512kbps download. I'm not sure this is correct > > as pppoa3 does use much less CPU (with 5% CPU, i can read a mpeg4 > > movie, and i'm _sure_ mpeg4 decoding is much more CPU intensive than > > simple memoves and usb urbs submissions) Is it normal ? > > No, it is not normal. How are you monitoring CPU usage? Just using > top, for example, won't show you any info about what's going on in the > kernel, because it all happens in softirq's. You need oprofile to do > that. Is it pppd that is using up the CPU? Do you have a uhci hc by > the way (the driver is inefficient, but that should be the same with > pppoax)? I also have an Athlon 2200+, and the speedtouch code comes > up pretty far down the list in my profiles (uhci-hcd is much more > present though). If you turned on verbose debugging, the logging may > be using up CPU. Try decreasing the buffer sizes (this is a module > parameter) - if that helps then you are being hit by the uhci hcd's > design problems. Also, if you are using multiple urb support (module > parameter) with a 2.4 kernel then turn it off - it has never been > reliable with 2.4. Which kernel are you running? Yeah i used top or visually looking at gkrellm (i know it's not a great monitor for this kind of thing) I use a 2.4.22-ck2 kernel with preempt and low scheduling turned on. And i'm using the kernel module in its simplest form as i let hotplug loading it up the way it wants and i don't have module entries in my module.conf concerning speedtch. -- Edouard Gomez Liste de diffusion modem ALCATEL SpeedTouch USB Pour se d�sinscrire : mailto:[EMAIL PROTECTED]
