BLU Meeting Notes: Wed, Mar 21, 2007
Linux Soup X: Red Hat 5 Virtualization
Written by Martin Owens

iSCSI --" Targets available for RH4U2, RH5, RH6 using a combination
of open/linux-iscsi;

Virtualisation, Utiliation normal hardware and configuration advantages;
Support for changing hardware around, maintenance of hardware and using
software that may not be quite suitable for the hardware platform; Use
less hardware by running more than one platform. Virtualisation requires
hardware extensions in the processor and host operating system.

1)Single Kernel Image (SKI), good and bad points where a multiple
vertualisation platforms share the same kernel. Speed advantages but at
the cost of flexibility
2)Full Virtualisation, Everything translated e.g. Vmware, mvs; the
software controls all the platforms, penalty of about 10% speed to
guests but the advantage is that the guests are not aware they are
running in a virtual environment and hardware incompatibilities are
mitigated by the vitiualisation. Host kernel runs in privileged
position.
3)Para-virtualisation, Each platform cooperatively talks to the
virtualisation system to improve performance. The down point is that it
wont work with unsupported guest platforms. e.g. Xen.

PAE --" older 32 cpu can only address 4GB of RAM, PAE allows you to
have more RAM by turning it into a paged system controlled by the cpu.
Can check the capability via /proc/cpuinfo or through intel website when
you are looking to prospectively buy hardware. Mostly used for
para-virtualisation. Older visions of Linux will support non PAE
systems.

A lot of laptop makers have hardware which support VTX/I (virtualisation
technology, x86/Itainium) but are required to be switched on in the
BIOS, unfortunately most don't provide a way to do
this. Some laptops provide  full VT support and this isn't required to
be switched on.

Paused migration where 1 guest is paused, moved and then re-enabled. So
long as both hosts have access to the same nfs disk the migration goes
well.

Live migration uses the current running system to copy over the guest to
another machine while running. The machine is brought to a lower
performance in order to finish off the move so as not to incur a race
condition with new memory and io updates on the old host server. There
is a small gap where the old guest is disabled and the new guest is
brought back. In effect a vidio could be running on the guest platform
while the migration is taking place and the video will continue to run
without stopping.

No real GUI yet for vitualisation when doing migrations, although much
is planned since all of the tools currently take advantage of cli tools
so the gui tools should be too far away. (compete with vmware
administration tools?)

(note) some fears that HV has a security flaw where the privilege
escalation allows possible root kits, new virtualisation machines should
probably be installed by the owner for secuity.

With the configuration file for the live virtualisation you can copy the
configuration to create a new virtualisation guest by changing minimal
parameters. (no tool yet to automate this task)

Hard drive sectioning can use and entire disk or a simple file. Simple
files allow LVM control including expanding the disk as required; 
allowing you to kick start a guest by over committing the disk the
beginning of the guest file is stored on (feature/bug).

When the guest crashes logs can be found in /var/xen/ allowing you
standard ways of understanding what may have gone wrong. When machines
are running in full the server can run a small good kernel in order to
dump logs to the system or even network options.

The limitations of the guest and host server software is not to do with
the software but the software support. it's possibles
to overcome marketing limitations with enough self support.

GFS -- clustering file system where virtual guests can access the
clustered file system. Better performance than nfs and includes locking
to stop each computer sharing the same space from corrupting each other,
gfs is more like a mountable disk that is elsewhere; nfs is more like a
file communication which is mounted like a partition but the requests
are not handled like disk io (maybe confused with iSCSI check).


New redhat version has some new tools that we request be shown,
including compvis 3D desktop.