Re: gkrellm eth0 chart
Bill WIlson <[email protected]>
| Newsgroups | gmane.comp.gnome.apps.gkrellm |
|---|---|
| Message-ID | <[email protected]> |
On Sun, 16 Apr 2006 07:14:19 -0700 "Joe Roback" <[email protected]> wrote: > gkrellm 2.2.7, FC5, kernel 2.6.16-1.2080_FC5, x86_64, intel e1000 ethernet card > > The gkrellm eth0 chart no longer displays a steady chart. For example, > I am transferring data at 100KB/s to a webserver. Instead of showing > 100KB/s, like it always use to, it shows 0k/s. then ~200k/s, then > 0k/s, then ~200k/s, etc. The chart is a bunch of spikes. I haven't > changed any settings and even tried removing my settings. > > Here is a screenshot. I have a steady 60kb/s upstream going. > http://www.roback.cc/tmp/gkrellm.jpg > > This even happens say transferring from my linux box to OSX using scp. > MenuMeters in OSX shows a steady 20MB/s transfer rate (gigabit > switch), but gkrellm is a bunch of spikes. It's an issue with the e1000 driver, see the thread: http://www.ussg.iu.edu/hypermail/linux/kernel/0308.0/0329.html And I'll insert here the email response I sent to someone else who asked me about this: -------------- inserted email: > Hi there. First of all, thank you for this great product - gkrellm. It's > always stays on my desktop. Now on point. This below is my short > conversation with Jesse Brandeburg from intel who is responsible for e1000 > NIC module/driver. Please read it and _please_ tell me any possible > solution to fix this... inconvenience. > Thank you! > > -----Original Message----- > [...] > Subject: strange behavior in e1000 driver in 2.6.16-rc2 > > Hi there. I am a regular linux user, and not developer. I have a poor > English but anyway I'll try to explain to you my observation about e1000 > driver (module) in 2.6.16-rc2 (specific this version). With 2.6.16-rc2 > my > network card and the e1000 driver works well, but I found out that the > traffic displayed in gkrellm (gkrellm.net) is... strange. The traffic is > something like this (my broadcast lan traffic for example): > 16k - 0k - 16k - 0k - 16k - 0k - 16k - 0k... and so on with period 1 > second. > With 2.6.16-rc1 with the same traffic in the same lan and same > everything, > the situation is like this: > 7k - 8k - 8k - 8k - 7k - 8k - 7k - 8k - 7k. In other words - the traffic > flows without interruption. It seems that in 2.6.16-rc2 the traffic is > going double in first second, then 0k, then again double, then 0k and so > on. > Because I'am not sure that my explanation is understandable I attach > some > gkrellm screenshots in one picture. This maybe will be more clearly. > My question is: is this a bug? Or some other "module" in kernel is > responsible for this? What is it and what provoked it? > Thank you! [ unfortunately I didn't quote Jesse's response where he simply stated this was intentional and for performance reasons] Hi, I'm afraid that if the kernel e1000 driver updates its statistics only once every two seconds then there is nothing that gkrellm can do to make the updates behave as before. Gkrellm can't set an update rate slower than once per second since this would require a redesign and it would adversely affect the other monitors. I don't know how much of a performance hit is involved but perhaps updating the counters once per second wouldn't be too bad. Your only option might be to change the kernel souce if you are compiling your own kernel. You would need to edit the file linux/drivers/net/e1000/e1000_main.c and change this one mod_timer line: mod_timer(&adapter->watchdog_timer, jiffies + 2 * HZ); to mod_timer(&adapter->watchdog_timer, jiffies + HZ); and that would get you to your once per second updates. Or it would be better acutally for data monitoring to go for at least twice per second updates with: mod_timer(&adapter->watchdog_timer, jiffies + HZ / 2); "jiffies" is a variable which has the total number of time ticks since boot, so you are setting the timer for "now + some_interval". Sorry I don't have a better answer for you. Bill -- http://gkrellm.net _______________________________________________ Gkrellm mailing list [email protected] http://ninja.linux-phreak.biz/mailman/listinfo/gkrellm