Blog of Ingo Jürgensmann

Today I discovered that one of my domUs at work is performing slow on its mounted NFS share. Bonnie++ and dd tests showed a network throughput of just 300 kB/s whereas the throughput was up to 110 MB/s from dom0 to NFS server. Some Google searches revealed that Xen has problems with NFS performance with non-standard rsize and wsize setting and especially with NFS over UDP.
Reducing rsize and wsize settings didn't help at all. The performance was still awful. After remounting the NFS share via TCP the performance was as expected. Such a huge difference in perfomance surprised me. Still, a strange bug in Xen, at least in Etch. Maybe it's already fixed in Sid?

After I arrived at home after work today I noticed that the machines at my parents home network were all down and just recovering from the outage. Two of my m68k autobuilders are located there. Nothing serious, I thought, because the machines are located behind DSL and maybe there was a short DSL outage.
Some minutes later I received two mails from my UPS that there was a power outage. Well, this explains the DSL outage as well, because the DSL is not attached to the USV (located in a different room).

Thu Jul 03 17:56:19 CEST 2008 Power is back. UPS running on mains.
Thu Jul 03 17:56:19 CEST 2008 Mains returned. No longer on UPS batteries.
Thu Jul 03 17:46:49 CEST 2008 Running on UPS batteries.
Thu Jul 03 17:46:43 CEST 2008 Power failure.

Ten minutes of power outage without machines being shutdown. It's a APC UPS BR800I with 3 machines (2x m68ks, 1x PIII@550), a SCSI hardware RAID with 7 disks and a 8 port switch hooked up to it. I'm really glad that I bought the UPS earlier this year - exactly for this purpose and to prevent filesystem errors on the buildds when there is a power outage. :-)

It's an ongoing drama: Flash player on PowerPC systems - you can either choose to ignore flash at all (no mozilla plugin installed) or the a non-working version of Gnash installed, which gives you no output of YouTube videos and hanging, CPU eating processes:

PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
16302 ij 20 0 63964 22m 13m R 75.0 2.2 4:27.56 gtk-gnash
16308 ij 20 0 44792 13m 10m S 5.9 1.3 0:24.46 gtk-gnash
16297 ij 20 0 54172 14m 10m R 2.6 1.4 0:11.07 gtk-gnash
16301 ij 20 0 54172 14m 10m S 2.6 1.4 0:10.92 gtk-gnash

I really doubt that there will ever be a working Flash player for non-Intel compatible architectures. :-(
I'll try swf-player next, but have little hope that it will be better than cr^HHgnash....

UPDATE:
swfdec-mozilla doesn't work properly either. It hangs iceweasel just a second after playing a YouTube video and the picture is broken.

Well, during the weekend some Debian machines were unreachable as MJ Ray wrote on debian-devel:

gluck, merkel, samosa and raff uncontactable (192.25.206.* network problem?)

I don't know anything more at this time, but wanted to push a small
message out so that others know it isn't just them and lists and IRC
are both still up, as far as I can see so far.

Except the topic on #debian-devel channel there was no other official notice, afaik.

I think handling of such issues can be improved. Although nobody apparently seemed to know what happened or when the problem will be solved, it would have been nice to publish information that actually were available:

1. What happened?


2. When did it happen?


3. What services are affected?


4. What is done to solve the problem?


5. When will the problem be solved most likely (ETA)?


6. Is there an alternative for the services affected by the problem?


7. Where can I inform myself about any progress?

A webpage similar to that one for DSA-1571 would be a very nice idea, IMHO.

AAAAA-was? Wer dieses Akronym nicht kennt, ist vielleicht noch nicht gänzlich verloren. VERA (Virtual Entity of Relevant Acronyms) löst diese Abkürzung laut dict wie folgt auf:

AAAAA
Anonym Association Against Acronym Abuse (telecommunication-slang)

Dagegen ist AbKüFi ja schon landläufig bekannt und heißt Abkürzungsfimmel. Beides sagt im Prinzip aus, daß man es mit Abkürzungen auch übertreiben kann, aber weder AAAAA noch Abküfi machen das so schön und kreativ wie es Jasper tut:


via http://www.aptgetupdate.de/index.php/2008/06/05/musik-generation-lol/

Sodele... der erste Arbeitstag in Berlin ist nun vorüber und es war recht anstrengend. Vornehmlich aufgrund des sehr warmen Wetters und der Tatsache, daß ich mit dem Fahrrad hin- und auch wieder zurückgeradelt bin. Pro Wegstrecke sind das rund 30 min Fahrzeit bei ca. 5.5 km Fahrstrecke, wobei ich das derzeit ohne Fahrradcomputer nicht so genau sagen kann.
Aber die Fahrtstrecke geht ein bißchen an der Spree entlang bzw. kreuzt diese mehrfach, am neuen Hauptbahnhof, der Waschmaschine aka Kanzleramt vorbei und auch die Dächer des Reichstags kann man ebenso sehen wie den Friedrichsstadtpalast. Mal vom Kopfsteinpflaster und den ganzen Autos abgesehen, eigentlich eine recht schöne Route, wenn man das in einer so großen Stadt sagen kann. Eventuell werde ich auch mal eine Alternativroute durch den Tiergarten ausprobieren, wenn das Wetter nicht mehr ganz so warm ist.
Aber derzeit fahr ich noch mit (geliehenem) Navi im Rucksack. Klappt ganz gut, wenn nicht grad lauter Verkehr an einem vorbeirauscht. Wird Zeit, daß ich mir dann mal einen eigenen kleinen Navi zulege. Wenn jemand einen Tipp für ein Gerät hat, daß gleichermaßen im Auto, beim Fahrradfahren als auch zu Fuß taugt, dann immer mal her damit! :-)

Russel Coker is looking for a Xen domU to rent for a new project he's going to start.
Using shared Xen hosting to lower costs is quite common these days. Steve Kemp does it and so do I (german). Prices are not that high that you can make a business out of this or earn your living from that, IMHO. At least not with a major investment beforehand. So, shared Xen hosting is mostly used to lower costs by sharing them amongst friends.

Just order a dedicated server, for example a dual AMD64, 4 GB RAM, 2x 400 GB harddisks, a /29 net and umlimited traffic (1 TB free at 100 Mbps, after that the bandwidth is reduced to 10 Mbps) for € 59.- and go for it. Get as much memory as you can get. I've found 2 GB RAM a little bit undersized, whereas I still have approx. 2 GB RAM unused/free for other domUs with 4 GB RAM, so there's some room for other domUs.
Currently I'm running 4 domUs:

1. 1 GB RAM: central services host (PostgreSQL, MySQL, Mailserver, IMAP/POP3)


2. 512 MB RAM: Apache


3. 128 MB RAM: providing shells for irc, Jabber, ...


4. 256 MB RAM: a VPN gateway for a local freifunk club

Anyway, when someone would consider sharing a dedicated server with some friends, go for it! It's better for the environment as well, when some people will share the servers instead each of them getting their own one.

Well, yesterday I spent most of my day to drive to Hamburg, deliver a new m68k (elgar.buildd.net) and relocate crest.debian.org and kullervo.debian.org to their new data center (same sponsor, but different data center location).

Everything went basically fine. Crest and Elgar needed some refitting of expansion cards, but that was expected and known. What was not expected was Crest doing some loooong fsck runs on its ext2/3 partitions. This lasted some hours and resulted in lots of stuff in lost+found. Sadly, XFS is not a real option for m68k buildds, I believe, although it works. But that showed me again why I don't like ext2/3... :)

Anyway, all three machines are working and accessible again. Thanks to NMMN for sponsoring and Andreas from NMMN for staying that long!

Als ich vor kurzem in Berlin zu einer Besichtigungstour für potentielle Wohnung war, blieb mir nichts anderes übrig, als in Berlin-Mitte das Auto zu parken und zu kapitulieren:



Sogar ein Ladenbesitzer konnte uns nicht sagen, ob man nun dort parken darf oder nicht. Der Parkscheinautomat wollte übrigens kein Geld annehmen.

Nachdem ich ja schon kurz nach der Bestellung bei Alice meinen Spaß hatte, nimmt das Drama nun weiter seinen Lauf:

3. Auftragsbearbeitung

Ihren Auftrag haben wir erhalten und bereits zur Bestellung an die T-Com weitergeleitet. Voraussichtlich Mitte Juni 2008 werden wir Ihren Anschluss schalten. Sobald uns eine Terminbestätigung der T-Com vorliegt, erhalten Sie noch eine schriftliche Bestätigung von uns.

Ja, wir schreiben das Jahr 2008 und man muss heutzutage fast einen ganzen Monat auf Telefon und DSL warten. Fantastisch!

Da ist die Überlegung dann nahe, von seinem 14tägigem Rücktrittsrecht Gebrauch zu machen und lieber bei der Konkurrenz lieber UMTS zu nehmen. Der Vorteil: überall verfügbar (Zug).

UPDATE: Dank "wundersamer" Umstände hat sich der Termin nun auf den 23.05. verschoben. :-)

Da ich ja durchaus schon früher über meine tollen Erlebnisse mit Alice-DSL berichtet habe, möchte ich das nun fortführen. Der Grund für das neuerliche Alice-DSL ist übrigens die Zweitwohnung in Berlin aufgrund des neuen Jobs in Berlin.

Naja, wie auch immer. Nach der Online-Bestellung gestern, bekam ich nun heute eine Email von Alice, daß noch die Daten des Vormieters fehlen würde, also Name und Telefonnummer und daß ich die fehlenden Daten im Kundenbereich im Web nachtragen könnte:



Das lustige dabei ist, daß oben steht, Name und Telefonnummer solle man nur eintragen, wenn man sie sicher kennt. Ansonsten solle nichts eintragen. Da ich weder Namen noch Telefonnummer vom Vormieter kenne, hab ich also brav nichts eingetragen und auf Weiter geklickt. Daraufhin kam dann der Hinweis hinzu, daß Name und Telefonnummer Pflichtfelder seien, also ausgefüllt werden müssen.

Testet Alice ihr Webinterface auch irgendwie mal?

Some days ago I complained about snmpd having a memory leak in Debian Etch, which seems to be related to the Etch version, as someone commented there. Anyway, todays complaint is against spamassassins spamd and its memory usage:

top - 18:36:15 up 30 days, 5:17, 12 users, load average: 9.60, 4.95, 2.38
Tasks: 170 total, 2 running, 168 sleeping, 0 stopped, 0 zombie
Cpu(s): 0.6%us, 1.4%sy, 0.2%ni, 31.7%id, 66.0%wa, 0.0%hi, 0.0%si, 0.1%st
Mem: 1048576k total, 1041264k used, 7312k free, 0k buffers
Swap: 1048568k total, 695652k used, 352916k free, 35568k cached

PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
31694 nobody 18 0 565m 360m 2396 D 1 35.2 0:08.45 spamd
23003 root 16 0 565m 344m 2456 D 0 33.6 40:54.15 spamd
27052 nobody 16 0 94216 40m 2456 D 0 3.9 0:29.96 spamd
9241 clamav 15 0 224m 39m 740 S 0 3.8 3:19.17 clamd
31711 nobody 15 0 89856 36m 2464 S 0 3.6 0:00.37 spamd
31750 nobody 16 0 89244 36m 2436 D 1 3.5 0:00.24 spamd
30751 root 15 0 87884 17m 2176 S 0 1.7 0:09.42 spamd

Some spamd processes are using a incredible amount of RAM, sometimes even eating up all memory including swap. I know that a bigger commercial site is suffering from the same problem as it was mentioned in their job interview with me lately.

Spamassassin is invoked by exim4 on my system during SMTP data checks with a condition like:

condition = ${if <{$message_size}{128k}{1}{0}}

So, this should ensure that only mails smaller than 128 kB are passed to spamd by exim. So, either that condition doesn't work (it should as it's basically the same as in the exim4 docs) or spamd is a memory hog even with small messages.
I wonder if there's any method to stop spamassassin from using that much RAM?

Something I already noticed in my previous company is the fact that snmpd is suffering badly from memory leaks.
For example:

PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
13904 snmp 15 0 856m 837m 2268 S 0 64.2 0:13.02 snmpd

This is from an amd64 xen system, whereas on an i386 xen host it looks like:

PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
3279 snmp 16 0 8064 1668 912 S 0.0 0.3 33:11.93 snmpd

Both are Etch systems, so same versions.

I've found Bug #421906 and my amd64 host is indeed using 2.6.18-6-xen-vserver-amd64 whereas my i386 is using 2.6.18-5-xen-686.
Does this mean that using 2.6.18-6-xen-amd64 would solve my snmpd problems, dear lazyweb?

Yesterday I signed the contract for a new job in Berlin. I'll start begin of June.
It's remarkable how fast the subscription to debian-jobs mailing list resulted in a new job, although I think that this is not necessarily the same for everyone subscribed there. :-)

Spring is upcoming, the trees are getting greener and everything is new and shiny - and so is Buildd.Net as well!

After a long time of being based on static html files, it's now based on a CMS that makes changes and additions more easy and offers other useful stuff. Next to come is some sort of SVN or git repositry...

So, be invited to visit and participate on Buildd.Net!