New updates build server

Vincent Danen

April 10, 2008

Just got my hands on a new updates build server. It's a quad-core xeon 2.13GHz with 8GB RAM. To put things into perspective, this machine will replace 3 different machines eventually... one to do the administrative stuff (AMD Athlon64 2.2GHz w/ 2GB RAM), one to build 32bit updates (dual core P4 3GHz w/ 1GB RAM), and one to do 64bit updates (AMD Athlon64 2GHz w/ 1GB RAM). Yes... I've been working with those machines for quite a few years now. Current kernel and OOo builds have definitely necessitated an upgrade.

To put things further into perspective, I compiled the Annvix kernel (2.6.22.19) on the two build machines and this new one (and also on my old trusty dual-opteron 1.8GHz w/ 2GB RAM that was my Annvix build machine... since I'm no longer working on Annvix, I was initially thinking of making that machine a new Mandriva updates build machine).

The 32bit build machine compiled it in 36:35.99m using 102% CPU. The 64bit build machine did it in 28:15.13m using 97% CPU. The dual-opteron did it in 23:47.04m with 152% CPU. This new quad-core xeon did it in 12:01.66m with 229% CPU. (Note that this is a single SMP kernel, not the multi-flavour kernels Mandriva builds).

I'm absolutely tickled pink by this. =) This will definitely make secteam more efficient.

Unfortunately, it also showed that 2008.1 will not boot on this system (at least with the 64bit Free DVD installer). 2008.0 had no problem booting and installing. Unfortunately, doing my urpmi --auto-select to upgrade from 2008.0 to 2008.1, it also looks as though 2008.1's kernel is to blame... the system is not booting at all now. =(

EDIT: using acpi=off makes it work, so the installer will run now (and the 2008.1 kernel boots, but the initrd is hosed giving the same error as 2008.0).

EDIT2: using acpi=off works, my quad-core has turned into a single-core. Definitely not cool. And acpi=ht doesn't work... it hangs. So according to the Ubuntu wiki, this means that "the issue is in the ACPI table parsing code itself, or perhaps the SMP code". Gustavo is telling me that there were some ACPI changes in 2.6.24, so that is most likely the culprit here.

/me sighs

Leave a Comment

Comments use MarkDown. Need help? MarkDown Cheatsheet