Script for automated testing of computer latency.

tommylight wrote:

Open FireFox, play a video, resize it several times, leave it playing.

Have we found anything that effects latency more than this combo?

Yes, copying large files, compiling software, rendering, video editing...
In short anything that uses IRQ requests will trigger latency excursions, hence video is a good scenario for testing.

here you asked for:

can you please add a way to find out the IP address and gateway address and ping the gateway say 50 or 100 times and report the times?
That would be helpful to point out some PC's having issues with new kernels and network cards causing loss of connection to controler boards.
So far Intel network cards are OK, some just need setting IRQ coalesing to 0, some work out of the box.
Realtek cards have something wrong in the kernel, and despite having good ping times tend to drop the link consistently.
Not sure how useful will that be due to above reasons, but maybe there is a pattern somewhere.

a) If this is not really relevant to latency, I might want to put it as an option in the initial menu. I might add a main menu item with label "Trouble shooting". It could be extended to look into other issues too. Would that be OK?
b) I am sure I can put the ping numbers settable with a default value of whatever you think is more reasonable.
Regarding IP address: the test candidate's or the gateways or both?
c) Is there a distinct kernel version where the problem started to occur?
d) Any other hardware than Realtek with same problem? I could test for this version/hardware and issue a warning pointing out to the very problem.
e) Guess this is for servo thread only?

besides:

f) copying large files,
g) compiling software,
h) rendering,
i) video editing...[Iquote]

f) is in the workshop, no ETA but the author is sweating
g) to i) users would require to install quite a bunch of packages way off making chips to do this. Therefore, I would prefer not to implement that, unless there is good reason to it all the same.

Comments welcome
greez
chris

seuchato wrote:

a) If this is not really relevant to latency, I might want to put it as an option in the initial menu. I might add a main menu item with label "Trouble shooting". It could be extended to look into other issues too. Would that be OK?
b) I am sure I can put the ping numbers settable with a default value of whatever you think is more reasonable.
Regarding IP address: the test candidate's or the gateways or both?
c) Is there a distinct kernel version where the problem started to occur?
d) Any other hardware than Realtek with same problem? I could test for this version/hardware and issue a warning pointing out to the very problem.
e) Guess this is for servo thread only?

A) Yes,
50, if the network is not ok it might take to long. IP is not important as a result, it is needed to find out if the network is setup and from that it is always easy to know the gateway, 192.168.0.nnn will have 192.168.0.1 as gateway, etc, usually that gateway is the router/access point.
C) Yes somewhere after 5.06 or 5.09 all the way to 6.02.
D) Not that we noticed. A warning would be OK, something like: even if the results of this test are ok there might be other limiting factors....
E) Yes
F) Thank you

P.S. It would be really nice if others would voice their thoughts on this, i am extremely spaced out lately, working on several projects every day not getting much sleep, so i might be sending you chasing my "fata morgana" .

Oh dang, the first smiley with glasses is not my doing, capital b and ) does that, apparently!

@seuchato
There is a new kind of latency with Mesa cards and that is network latency which is biting on later kernels, particularly where realtek NIC's are used. Refer the later pages of  forum.linuxcnc.org/27-driver-boards/4691...nishing-read?start=0
Its pretty clear now that this is a debian issue with the drivers. Compiling the kernel or using a non Debian based system like ArchLinux seems to solve the issue.

I've reported this as a Debian bug. 
bugs.debian.org/cgi-bin/bugreport.cgi?bug=1022170
If anybody has more feedback, it would be great to post the info on this bug report. Steffan Moller is going to make sure this is seen by the right pople and has assured me it will be fixed. A bit more support might help...