[Nfd-dev] NDN-RTC: NFD processing logs
jburke at remap.UCLA.EDU
Sun Feb 8 11:09:32 PST 2015
I have opened Bug #2492 for this.
From: Jeff Burke <jburke at remap.ucla.edu<mailto:jburke at remap.ucla.edu>>
Date: Sun, 8 Feb 2015 18:53:14 +0000
To: Junxiao Shi <shijunxiao at email.arizona.edu<mailto:shijunxiao at email.arizona.edu>>
Cc: "nfd-dev at lists.cs.ucla.edu<mailto:nfd-dev at lists.cs.ucla.edu>" <nfd-dev at lists.cs.ucla.edu<mailto:nfd-dev at lists.cs.ucla.edu>>
Subject: Re: [Nfd-dev] NDN-RTC: NFD processing logs
I don't agree – unless the NFD team is aware of a change in the codebase that would potentially fix this problem, and you can selectively apply only those changes as a patch, to see if they fix the problem, simply upgrading provides no information on the root cause. Thus, we may eliminate the symptom without understanding where it comes from or knowing when it might come back.
Given our goal of having this system work "well enough" to use in real meetings, etc. and to do large-scale demos with WUSTL, simply eliminating the symptom is insufficient.
Simplest way to find the root cause is to debug in the current, deployed system.
A hybrid approach is to substitute an NFD 0.3 forwarder (on a different, identical box) as the hub and see if fixes the problem. This might inform which commits to apply/roll back in the testing process. But, somehow, we should try methodically to find out why this occurs.
We can provide additional hardware configure per the above, if you'd like. But we need the NFD team to engage with debugging the current running code on the hub. This will be more productive and more likely to get to the root cause than having us make further changes to the setup.
From: Junxiao Shi <shijunxiao at email.arizona.edu<mailto:shijunxiao at email.arizona.edu>>
Date: Sun, 8 Feb 2015 11:42:25 -0700
To: Jeff Burke <jburke at remap.ucla.edu<mailto:jburke at remap.ucla.edu>>
Cc: "Gusev, Peter" <peter at remap.ucla.edu<mailto:peter at remap.ucla.edu>>, "nfd-dev at lists.cs.ucla.edu<mailto:nfd-dev at lists.cs.ucla.edu>" <nfd-dev at lists.cs.ucla.edu<mailto:nfd-dev at lists.cs.ucla.edu>>
Subject: Re: [Nfd-dev] NDN-RTC: NFD processing logs
In any case the delay isn't in forwarding pipelines or strategy according to logs.
@Davide is expert on faces and related libraries, and he agreed to determine whether the problem is in UdpFace or a lower layer.
Debugging this problem doesn't depend on new Ubuntu port.
However, upgrading NFD to latest git-HEAD version is necessary for bug reports, because this problem may have been fixed already and it's wasting time trying to debug an older version.
After upgrading NFD and confirming it still exists, please officially file the bug on NFD Redmine site.
On Sun, Feb 8, 2015 at 11:33 AM, Burke, Jeff <jburke at remap.ucla.edu<mailto:jburke at remap.ucla.edu>> wrote:
Junxiao & all,
I want to strongly advocate for debugging the actual, running system that we have now, rather than waiting for new code and using another setup that may or may not exhibit the same problem.
Would it be possible to work with the system in situ with remote access rather than waiting on the ubuntu headless build? (Alternatively, I would rather host you at UCLA for a week, or fedex you our Mac test machines, than write new application that just attempts to replicate the problem.
What Peter was referring to is an Ubuntu port of ndnrtc that will generate the same type of traffic, not a simulator. We need this for testbed deployment. However, this work is not done yet, and we are trying to move quickly to real world use, and I really don't think we should make this dependent on further app development work.
On Feb 05 we identified a few possible places that can cause delay:
* C++ memory allocator
* UdpFace or its base class DatagramFace
The logs you gathered appear to be complete according to what we requested.
@Davide, can you gain some insights from the logs?
Regarding the 30-second interval pattern discovered by John: the only 30-second interval I know of in NFD is UDP face idle_timeout, when this option is unset in nfd.conf.
Every 30 seconds, UdpChannel will enumerate all its faces, and close all faces that haven't received anything. There's no evidence that any face is created, because FaceIds are consistent throughout the logs. However this scheduled event will take some CPU time, but I don't believe it can cause a delay of more than 5ms.
To rule out this possibility, adjust face_system.udp.idle_timeout option in nfd.conf (to "50"), and test again.
>From an earlier log snippet, I notice that the NFD version you are running is older than September 2014.
Please use git-HEAD version for any bug reports.
As you told me on Feb 05, you will create a traffic simulator that can have same traffic pattern as NdnCon that can run on headless Ubuntu servers.
I'll be able to test on controlled environment when that tool becomes available.
On Fri, Feb 6, 2015 at 4:14 PM, Gusev, Peter <peter at remap.ucla.edu<mailto:peter at remap.ucla.edu>> wrote:
So I’ve run setup today and gathered logs.
I’ve setup NFDs with the following settings
And then I run tcpdump on each machine like this:
$ sudo tcpdump -w ~/dump.pcap -i en4 udp port 6363
Here is the summary on “problematic” packets:
All the log files and tcpdumps you can find here:
Thanks for all the help!
Programmer/Analyst @ REMAP, UCLA
peter at remap.ucla.edu<mailto:peter at remap.ucla.edu>
+1 213 5872748<tel:%2B1%20213%205872748>
_______________________________________________ Nfd-dev mailing list Nfd-dev at lists.cs.ucla.edu<mailto:Nfd-dev at lists.cs.ucla.edu>http://www.lists.cs.ucla.edu/mailman/listinfo/nfd-dev
_______________________________________________ Nfd-dev mailing list Nfd-dev at lists.cs.ucla.edu<mailto:Nfd-dev at lists.cs.ucla.edu> http://www.lists.cs.ucla.edu/mailman/listinfo/nfd-dev
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Nfd-dev