[Gllug] Crashes (owing to the heatwave?)

[David Damerell]

On Friday, 8 Aug 2003, Jon Masters wrote:
>The question I had more specifically related to how the S390 handles
>parititions as in mainframes such as this it is possible to create virtual
>machine partitions and run a different Operating System image within each.
>Sun Microsystems are also beginning to use this concept more and more.
>There are various technical issues with removing a CPU during runtime and
>I asked as I would like to know how it is done in the S390/Linux case.

As I understand it [1], on the S/390 - unlike the partitioned Sun
machines - the hosted operating systems (I forget IBM's terminology
for them) are not tied to specific hardware. On a Sun E10K or later
machines, a partition is running on a definite subset of the CPU
boards, and although CPU boards can be moved between partitions this
is flaky, and a partition's first CPU board cannot be reassigned -
also peripherals are attached to CPU boards making this still more
problematic.

Conversely, on the S/390 the entire machine is potentially available
to every hosted OS; this means that CPU and memory can be more readily
reallocated between them (ordinarily automatically within predefined
limits), that you can have many more hosted OSes than physical CPU
boards or whatever (Telia have c, 6,000 Linux installs on one S/390),
and that any given piece of the machine can be taken out of service
completely (there's a funky "memory RAID" going on such that the
failure of one piece of memory doesn't cause immediate problems, and
the physical component can be identified as dead) without adverse
effects, powered down, and replaced.

[1] I've never adminned an S/390, this could be a total lie.

-- 
David Damerell <damerell at chiark.greenend.org.uk> Kill the tomato!

-- 
Gllug mailing list  -  Gllug at linux.co.uk
http://list.ftech.net/mailman/listinfo/gllug