Showing headlines posted by dba477

Notice that installation Ubuntu HVM and runtime performance at opensolaris 2008.11 Dom0 is extremely low versus same HVM perfomance at Xen 3.3.1 openSUSE 11.1 Dom0 on the same box with C2D E8400 and 8 GB RAM. Actually, Linux as well as Nevada HVM DomUs demonstrate very low performance even been installed on zvols. Paravirtual DomUs behave at opensolaris 2008.11 absolutely normal been installed on zvol’s images.

My primary target was testing xVM performance at opensolaris 2008.11(based on 101b Nevada build). Solaris Nevada works smoothly with images located on UFS and zpools.Opensolaris 2008.11 requires only zpools as image devices for Xen Guests to be installed in reasonable time.I believe it's a known issue caused by ZFS. From my standpoint the most impressive UI feature is Package Manager. It's really helpful.

During installation phase we would have to drop to shell and undertake some special actions to prepare runtime profile.Pygrub won’t be able to load DomU utilizing ZFS based image. In particular, ZFS-ID of boot environment will be obtained to compose extra line in runtime profile.

Install Xen Server package and booting openSUSE 11.1 RC1 Dom0 went pretty smoothy, however attempt to start vm-install either via YAST command center or from command line just brought up frozen blank window.

Failure to virt-install F10 PV DomU and repeated issue with CentOS 5.2 in stub domain after same port via http://gitco.de ( upgrading only hypervisor to keep virt-install alive at Dom0 ) brought me to idea just go through traditional “hg” cloning mercurial repository at http://bits.xensource.com and build local Xen with 2.6.18.8-xen kernel

To create loadable Fedora 10 PV image we would have to manage at Xen 3.3 CentOS 5.2 Dom0 either multibooting with Hardy Dom0 or remote.To install Fedora 10 PV DomU local NFS share will be utilized. Local Apache Server simulating HTTP mirror may be used as well. Installer provides both options. Attempts to perform install on Ubuntu Hardy Dom0 failed. Looks like Hardy has problems with the most recent Fedora’s file system. ISO image mounted via losetup doesn’t work correct versus it happens on CentOS 5.2

As of November 11 Intrepid Xen 3.3 Hypervisor may be back ported to Hardy Dom0 via repositories. Following bellow is a brief instruction for installation from scratch Ubuntu Hardy Dom0 with 2.6.24-21-xen kernel running with Xen 3.3 hypervisor

This is potentially a more time consuming backport but with Xen 3.3 being a huge improvement over 3.2 it would definitely help Ubuntu server market share to have Xen 3.3 available in Ubuntu 8.04. Many server operators will want Xen 3.3 but will want to continue running the LTS release rather than upgrade to Intrepid.

To load Ubuntu Intrepid Server PV DomU via serial console "pvops" kernel and ramdisk usually gets copied to Xen 3.3 Dom0. Parameters specifying root file system at image device and hvc0 console are also supposed to be included into startup profile at Dom0. At the same time all mentioned parameters may be specified via new entry into /boot/grub/menu.lst located at DomU, been placed into corresponding entry kernel line of grub configuration file at DomU. File /etc/event.d/tty1 at DomU should be also modified to work for xen console instead of vfb. All written above is not CentOS 5.2 specific, because pygrub comes to the system with Xen 3.3 port , no matter is it Debian or Gentoo

Using Ubuntu as a Xen guest is now a supported option included in the standard server kernel and is a choice when building virtual machines with python-vm-builder. Install Ubuntu Intrepid Server HVM DomU and create PV DomU profile to load image has been created.

This posting mostly deals with utilizing same image file for loading HVM and PV Ubuntu Hardy DomUs at SNV_98 Dom0. Due to ongoing issue with "rge0" driver Solaris Nevada (98) causes failure to run “apt-get install” at Ubuntu (Debian) PV DomUs on a wide range of ASUS (P5E3,P5K),GigaByte,MSI boards,having RTL 8110SC integrated as second NIC, until checksums offloading at PV DomU will get disabled via “ethtool” ( for instance) and provides a workaround to get Gnome Desktop working stable at Ubuntu PV DomU no matter on Xen Release (>=3.1.2) and OS (Solaris,Linux) running at Dom0.

Attempt to reproduce for OpenSuse 11 procedure worked so smoothly for RH’s PV DomUs , described in previous post, appeared to be a bit more complicated. Pygrub installation profile reading data from HTTP source at Dom0 still worked fine, but Xen 3.2 Ubuntu’s 8.04.1 pygrub failed to read image had been created for OpenSuse 11 PV DomU (/dev/sda11) . Just an ordinary “xm” profile was required to finish configuration and load PV DomU when image was already created on block device, say /dev/sda11. Xenified kernel and ramdisk had to be copied of image device to Dom0 file system. There may be several ways of achieving this goal.

Due to known bug virt-install was broken on Ubuntu 8.04 since day of it’s GA and never has been fixed as far as to my knowledge. However, xen-enabled ISO’s like RH’s ones allow to perform PV Guest install via pygrub and repository placed on Apache Server running at Dom0. Notice, that virtual frame buffer would work fine for CentOS PV at Ubuntu Hardy Dom0 (2.6.24-21-xen).

The procedure down here has been developed due to virt-install failure to handle OpenSuse 11 PV DomU install at Xen 3.3 (3.2),(3.1) CentOS 5.2 Dom0 either via NFS or HTTP shared directory. Perform initial connect to Apache Server running at Dom0 , deployment and configuration (until first DomU reboot) via profile openSUSE11PV.cfg When OpenSuse DomU first time goes down for reboot, shutdown Suse DomU via “xm” and switch to pygrub profile for loading scp'ed images created via vm-install at OpenSuse 11 Dom0.

It’s well known that OpenSuse’s utility vm-install fails to perform RH PV guests install via NFS shares. Install from Internet repositories requires software download about 4GB of data for any CentOS 5.X of Fedora PV DomU. Corresponding ISO images have been copied to CentOS 5.2 box running Apache HTTP Server and located on same LAN. Then this images were loop mounted to corresponding folders under /var/www/html. This procedure provided an option for vm-install (via http) RH PV Guests at Xen 3.2.1 OpenSuse 11 Dom0 in pretty short time frame about 40-50 minutes for every DomU been installed.

Posting bellow describes setup Ubuntu Hardy PV DomU via attacment block device to Ubuntu Hardy HVM DomU at the same CentOS 5.2 Dom0. Standard debootstrap procedure has been used to create initial image on this device. This step allows to avoid “scp” image files from remote Ubuntu HH box to Dom0 and use /dev/sda(X) with prepared image immediately on the same box,referencing the same device from Ubuntu Hardy PV profile.

Port Xen 3.3 to CentOS 5.2 has been done via http://www.gitco.de/repo/xen3.3.0 . Install CentOS 5.2 PV DomU via virt-install utilizing NFS share with mounted loop DVD ISO image. Create profile PV DomU as follows not using vfb at this time and setup “vnc” at DomU afterwords.

Debian Etch install took about 2 hr on the box with C2D E8400, 8 GB RAM built up on P5K Premium/WIFI board. South Bridge ICH9R has been setup to AHCI mode with 2×250 SATA drives attached. dom0_mem was set to 4GB in /boot/grub/grub.conf. Debian HVM allocated 2 GB memory , Ubuntu 8.04.1 HVM 2.5 -3 GB memory . In case Ubuntu HH Desktop stub domain install it’s important to have enough memory on the box for Dom0 and DomU.

Working through weekend brought some positive results. However, kernel = “/usr/lib/xen/boot/ioemu-stubdom.gz” and device_model = ‘/usr/lib/xen/bin/stubdom-dm’ for HVM Domains show slow performance vs traditional approach for HVM DomUs creating at Xen 3.3 Linux Dom0s.

We use the most recent Xen 3.3 port to CentOS 5.2 via http://www.gitco.de/repo/xen3.3.0. Prepare grub's /boot/grub/menu.lst at DomU side for pv-grub booting.