Arrfab's blog Some tips and tricks, mostly around CentOS

Fosdem is over .. and it was a good edition. From a CentOS perspective, i'd say that the booth was better than in 2007 and because we were not at the same place, i think more people had a look at the booth ... I was at the booth and/or in the devroom for all the CentOS talks (i've posted some pictures) but i had opportunity to at least follow some other talks :

• Wouter Verhelst's talk about Belgian eid on Debian (that reminds me of course my 2007 talk ...) ;-)
• Alasdair Kergon's talk about new features in LVM2
• Jens Kühnel's talk about SELinux (Jens is really a good guy to talk to ... always a pleasure to see/talk with him)

Let's go back to normal life ... and try to prepare next year .. ;-)

What can i say ? hmm ... I've posted some pics online here so that you can already feel the Fosdem event mood ...

I gave the 'Introduction to CentOS' yesterday and it's scheduled today too ... (30 minutes to go ....)

I can already say that i enjoy the Fosdem event .. cool organizers, nice place to be (but am i objective, being a belgian guy myself ? ;-) ) and of course it's always good to see people you're chatting with or sending mail to .. but in the real life ...

We share the devroom with Fedora and our booth is next to them ... and , to quote what they say on their wiki about us, "they are nice people" ..

More informations/pictures to come ..

It's with a great pleasure (again ..) that i'll be at the Fosdem 2008 event with folks from the CentOS project .. We'll have a booth and a devroom that we'll share with Fedora (like last year) so if you just want to tell us what you think of CentOS, what can be improved (there is open debate scheduled in the devroom) or just to say 'hello' and discuss with some other CentOS community members , feel free to come ...

And because it seems mandatory this year, here is the famous logo that i forgot to put in my blog

Because the company i'm working for is a Citrix partner and that Citrix wants us to become Xen partner (since Citrix acquired XenSource several months ago), i decided to download and test their XenServer on a IBM HS21 quad-core. Funny that to administer Xen graphically you need to use their .Net application on Windows .. and no Linux console available ! (while i remember that on previous version a java and so cross-platform version was available).

But the thing that surprized me more is that they are using CentOS packages for a lot of RPMS on the Dom0 ! In fact , i wanted directly to ssh in the xen box to see what was inside and `uname -a` answered me : Linux xen 2.6.18-8.1.8.el5.xs4.0.1.125.163xen #1 SMP Mon Aug 13 09:27:46 EDT 2007 i686 i686 i386 GNU/Linux ... That sounded very familiar to me ... so from where does the rest of the system come ? : rpm -qai|grep "Vendor: CentOS"|wc -l : 185 (out of 220 packages) ..

In fact, even the standard CentOS-Base.repo and CentOS-Media.repo are still in /etc/yum.repos.d/ (but they are modified to exclude kernel and xen packages …

I had recently to import a LVM previously created on a SLES9 (in fact i scratched the SLES that was installed on a separate disk). Of course lvm is lvm so i had direct access to the datas sitting in my logical volume. Everything was fine except that i was asked in the same time to extend the filesystem (a new disk was added in the volume group). Because i told the customer that it was easy with resize2fs (now supporting online extend on el5 , while you needed ext2online on el3 and el4), i decided to do it directly after my migration (understand after having scratched and replaced the SLES). But i had a surprise when trying to extend the filesystem : it didn't work !

resize2fs answered me : 'resize2fs: Operation not permitted While trying to add group' .. and /var/log/messages told me : 'No reserved GDT blocks' ... hmm, what does that mean ? inspecting the filesystem with `tune2fs -l` showed me that the ext3 filesystem created on SLES was lacking an important feature (already present for a while on RHEL/CentOS/others ...) : the resize_inode function was missing in the Filesystem features .. damn ... `man tune2fs` was not a great help because it seems …

Usually i deploy mainly IBM servers that use broadcom network cards and i swear i've never encountered any problems with such cards. The other day i was asked to setup a small pc with two sata disks (in raid1). I was sure that CentOS 5 could handle the integrated network card (from `lspci`: Ethernet controller: Marvell Technology Group Ltd. 88E8056 PCI-E Gigabit Ethernet Controller (rev 12) ) because i already had installed exactly the same box with 5.0 . Of course i tried to setup 5.1 and it was not possible to use that integrated card anymore. Frustrating especially when i wanted to setup the box over the network, like i always do ...

From the 5.1 release notes, it was stated that the sky2 module was updated to a newer version (version 1.14) but i was not expecting that it removed hardware compatibility ! In fact, the sky2 module was backported by Red Hat into 5.1 but it was decided in the main kernel tree to not support anymore the Marvel 88E8056 chipset with the sky2 module .. at least that's what i understand from this mail sent by Linus himself (search for sky2) . The only way to get …

When Red Hat released 5.1, everybody wanted to test a new kernel parameter that could adjust the system clock rate at boot time to something else than the standard 1000Hz clock rate. A lot of testings has been done by the CentOS QA Team and you can see the results here : http://bugs.centos.org/view.php?id=2189 (Notice that Xen guests don't need the system clock rate to be modified because they already have a 250Hz kernel)

As you can read at the bottom of the comments, it seems there was a typo

in the [official RH Release

Notes](http://www.centos.org/docs/5/html/release-notes/as-x86/RELEASE-NOTES-U1-x86-en.html)

you'd have to read divider= and *NOT* tick_divider= !

It seems so to work with the correct kernel parameter and so there is no need to build a kernel-vm for CentOS 5.1 guests .. (it's still needed for example for 4.x ..). The CentOS 5.1 Release Notes have been corrected to reflect the good divider= option .

Here is a small benchmark of an idle minimal CentOS 5.1 i386 running in a VMWare-server guest VM without (before 20h40) and with the divider=10 option (after 20h40 , so …

Ok, nothing linux/centos related, but still good to see ... http://www.tetesaclaques.tv .. real stupid fun .. Attention : in french only

I recently had to setup CentOS 5.1 x86_64 on a remote Hetzner dedicated server . CentOS is not listed in the supported distributions but that's not a problem, especially if you have already played with the remote vnc installation mode. The only problem is that their dedicated servers have a newer Realtek Gigabit controller that is not supported by the CentOS 5.1 kernel (from `lspci`: RTL8111/8168B PCI Express Gigabit Ethernet controller ). But the good news is that a driverdisk is available on the CentOS Wiki.

I choosed to setup a minimal Fedora Core 8 (supported through their web control panel) and from there modify grub to launch the CentOS 5.1 x86_64 setup. I was thinking writing the 'step-by-step' procedure somewhere on the wiki but i've found that such procedure (even if i didn't follow it completely because i installed first and quickly a minimal Fedora core 8 ) was already written by someone else on the Hetzner Wiki . My advice : add the 'noipv6' parameter to speed up the installation and avoid anaconda trying to get a ipv6 address through dhcp (see my previous note)
The other good news is that they have an internal CentOS 5 mirror so …

Some of my customers are running RHEL 5.x  on ppc64 (IBM Blade JS20/21 , OpenPower, System I or System P) but it was frustrating that RPMForge had no ppc build .  And there is also a plan to release CentOS 5.x for ppc/ppc64 arch. So i decided to rebuild (through Mock) all the RPMForge srpms on a Mac G4 . Packages rebuilt so far are now in testing, and the list can be seen here.

Hope to receive feedback from users using el5 on ppc/ppc64 .. When all srpms will be rebuilt, i plan on (probably) using the G4 to rebuild as well for fc8ppc.

Packages will normally appear on the official RPMForge mirror when everything will be stabilized ... and that can take some time because building on a (old) G4 400Mghz is not that fast ... (hardware donations accepted ;-) )