When faster WiFi means unusable connection

I recently moved home and got FTTC with PlusNet, the speed is good when measuring (almost the advertised 80Mb/20Mb) but the connection was unusable due to TCP connections hanging every few minutes (very annoying with ssh but screen helps, worse when using a website for a payment and needing to retry and trust you will only be charged once).

Yesterday I decided to sit down and investigate. Router has logs which were quite helpful. A lot of things like OUT: BLOCK [9] Packet invalid in connection (Invalid tcp flags for current tcp state: TCP [192.168.1.73]:54426->[46.19.168.229]:443 on ppp3)

This followed the laptop being seen moving from interface ath10 to interface ath00 and it was moving back and forth quite often.

Looking at the logs on one of the laptops those switches looked like wlan0: disconnect from AP b8:d9:4d:41:76:fb for new auth to b8:d9:4d:41:76:fa

What happened is that default settings on PlusNet router is to have “identical” 2.4GHz and 5GHz networks so the devices believe they are the same network and switch between AP, but they are actually different and the connection tracking gets reset each time such switch happens.

Disabling the 5GHz network made my connection usable, I could probably just change its settings to make it separate.

Success

In November 2012 I started running an irregular rebuild of all Mageia packages on x86_64, discarding the built packages, to just detect build breakages.

At first it was running a few times a month, now once a week, except before releases where I run it twice a week.

For the first attempt, on 2012-11-28, we had 10949 packages and 1104 failed to build (10%), by the time of the second attempt 3 weeks later we were down to 6.7% of build failures.

Since then, the distribution has been growing and this has helped detecting packages that needs to be fixed early, really helping for the mass rebuild of each release.

• Mageia 3 was released on 2013-05-01 with 133 failures out of 11008 (1.2%), most of them being random failures due to the use of make -jN.
• Mageia 4 was released on 2014-02-01 with 24 failures out of 11739 (0.2%)
• Mageia 5 was released on 2015-06-20 with 42 failures out of 12455 (0.2%)
• Mageia 6 is going to be released on 2017-07-XX with 0 failures out of 13650 (0%)!

For those interested in the technical details, I am using iurt (the same as we use on Mageia build system) to rebuild everything, creating a new chroot for each package.

I build 16 packages at once, with -j4, on a virtual machine having 32 cores and 200G ram which I use as tmpfs for the builds.

A full rebuild takes about 20 hours.

Résultats de l’élection présidentielle française 2017 à Londres

Pour une raison inconnue, le site du ministère de l’intérieur ne fournit pas de résultats par commune pour les français de l’étranger, contrairement à la métropole.

Les données étant fournies sur data.gouv.fr mais pas très lisibles, je les avais extraites. Je me suis dit ce matin que ça pourrait intéresser d’autres gens, donc les voici.

Résultats au premier tour

	Nombre	% Inscrits	% Votants
Inscrits	93647
Abstentions	49721	53.09
Votants	43926	46.91
Blancs	157	0.17	0.36
Nuls	218	0.23	0.5
Exprimés	43551	46.51	99.15

Nom	Prénom	Voix	% Inscrits	% Exprimés
MACRON	Emmanuel	22385	23.9	51.4
FILLON	François	10796	11.53	24.79
MÉLENCHON	Jean-Luc	4934	5.27	11.33
HAMON	Benoît	2910	3.11	6.68
LE PEN	Marine	1225	1.31	2.81
DUPONT-AIGNAN	Nicolas	535	0.57	1.23
ASSELINEAU	François	288	0.31	0.66
POUTOU	Philippe	194	0.21	0.45
LASSALLE	Jean	122	0.13	0.28
ARTHAUD	Nathalie	91	0.1	0.21
CHEMINADE	Jacques	71	0.08	0.16

Résultats au second tour

	Nombre	% Inscrits	% Votants
Inscrits	93672
Abstentions	46805	49.97
Votants	46867	50.03
Blancs	828	0.88	1.77
Nuls	205	0.22	0.44
Exprimés	45834	48.93	97.80

Nom	Prénom	Voix	% Inscrits	% Exprimés
MACRON	Emmanuel	43629	46.58	95.19
LE PEN	Marine	2205	2.35	4.81