Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[burakkucat]

It is good news that you have solved the mystery.

It is bad news that it has cost you at least £900 to do so.

Your next step must be to report it to WatchFront (t.a. Firebrick) and see how they respond. A credit of £900 to your bank account would be an appropriate gesture.

[Weaver]

The original thread at
    http://forum.kitz.co.uk/index.php/topic,17364.msg317505.html#msg317505
has the graph showing the onset of the phenomenon.

[Weaver]

Here is a dripping blood CQM graph for today's tests, showing line 1, initially on port 1.
    https://flic.kr/p/H6DXj5
The bright red coming down from the top of the graph, so-called dripping blood, represents loss of PPP LCP echo requests or the replies. See earlier post for timings linked to the changes in activity. The dripping blood is finally cured and disappears from the graph.

[Weaver]

Recap: this is how it all started, here's the original CQM graph showing the onset
    http://forum.kitz.co.uk/index.php?action=dlattach;topic=17364.0;attach=19619;image

I think that was the evening of 2016-03-24.

[Weaver]

It could be that this was a regression introduced in 1.39.00 (from 1.38.001) released on 2016-03-20 because the release notes mention work done relating to (i) PPP LCP (!), and (ii) PPPoE. What I don't understand though is the timing. When the firebrick auto-installed 1.39.00, I do not know. It's set to auto-upgrade to new releases but I don't know when/how it chooses to install them. I don't have the Firebrick logs for that day, unfortunately. But why it suddenly started showing the problem at around 19:15, I don't know. If it were 02:00-ish then that would be a plausible time for a reboot after a download.

[burakkucat]

Why not just gather all the relevant information together and then send the bundle to Adrian Kennard? I'm sure he will respond and investigate.

[Weaver]

I'm emailing RevK via the support dept at A & A, which is how he prefers it, so that all bugs and problems are recorded in one place. And I'm referring them to this thread and sending them the relevant CQM graphs, as I don't think the graphs in the earlier thread are accessible as they were attachments, which can only be seen by logged-in visitors, I believe.

[burakkucat]

<snip> as I don't think the graphs in the earlier thread are accessible as they were attachments, which can only be seen by logged-in visitors, I believe.

Yes, that is correct. Attachments are only visible to logged-in members.

[burakkucat]

Just wondering . . . Has there been any feedback (or developments) on this issue? 

[Weaver]

I reported the bug to AA, but they simply denied that there's any possible problem. this was despite a very detailed bug report and rationale, with a link to this thread.

So no luck at all. it's possibly my fault that I just sent the bug report to the support staff I had been dealing with, as opposed to RevK direct, maybe the latter would have helped my cause.

If another customer might be able to reproduce this, then that might help a lot. especially given the fact that the current reported is decidedly only half with-it.

It seems to me that nothing special is required to reproduce this. Just a Firebrick with the right (currently on release) firmware load, and the use of port 1 for a PPPoE device as opposed to for the LAN. I think the default/example config starts users off using port 1 for the LAN, so covering up the bug, if I am right. As an accident of history, I had always been using port 4 for the LAN and ports 1-3 for PPPoE modems. I realise that I don't know if multiple modems are _required_ in order to see the bug. I have never tried it with one modem.

If anyone is willing to be a guineapig, then I could lend them a Firebrick and a suitable modem,  seeing as I now have two. There would have to be second, independent (and forceful) bug report.

[burakkucat]

That is very unfortunate and rather surprising that it was dismissed without any obvious investigation.    Regardless of the above, I still think you should directly involve Adrian Kennard.

As for having someone else attempt to reproduce the problem, it would have to be an A&A user . . . 

[Weaver]

Yes, I suddenly realise that someone else would need to see the PPP LCP pinging service to see the bug. There is someone else who is running a server that provides a similar service using a firebrick 6000 series ping box over the Internet, see
    http://www.thinkbroadband.com/ping

That might, or might not, be enough to demonstrate it. The problem is that I don't know how many timing-related issues come into this. It may be that the network connection speed and latency has to be "just so". Who knows. there are no guarantees though.

If I were looking into it, I would start simply by code inspection, in a spirit of optimism, given that we know when it was introduced, and I see that the change log gives a few ideas of mods that might be relevant. The port 1 (C - prob 0?) thing has to be a dead giveaway, but I can't for the life of me thing how it's come into it. Of course, zero is special, but it's really surprising to me.