Being the guilty party (and in the interest of forestalling Yet Another
Round of TP:C Discussion), I guess I'm the right person to respond.

> There is a big gap between hundreds and thousands.  My question ...
> is zVM (with up to 16 engines) really capable of supporting
> thousands/ tens of thousands of servers in the real world,
> and IBM just being coy?  Or is it really hundreds?  If it is
> in the thousands, is this theoretical, or are there really
> customers using it at this level?

You have the critical point in the next paragraph. How many guests you can
get is *entirely* dependent on the characteristics of the workload.

> I recognise that this question is dependent on a wide range
> of variables.  These would include the fact that the actual
> workloads encountered will matter a great deal.

TP:C was a very specific workload and a very specific attempt to determine
what was *possible*, not necessarily what was *practical*. Given that
there's realistically no such thing as a "standard" business workload, the
only possible answer is that "it depends". Can you get 41,400 server
instances? Yes. Do they do useful work? Not much, really, other than to
demonstrate the viability of my conception of mass production use of
virtualized infrastructure under some pretty nasty edge cases.

A more practical case is likely to be in the small end of thousands,
ASSUMING (note emphasis) the workloads are relatively spiky and have long
quiet intervals between spikes AND that there is sufficient memory and
paging space to support the demands of the applications. Workloads that
remain constant and substantial over time will limit the scalability of the
solution. The z890 and z990 help by relieving some of the physical resource
constraints, but there's still only so much CPU to go around, and a lot of
"modern" applications waste enormous amounts of physical resources simply
because they *can* on other platforms -- garbage in, garbage out.

If the workload varies in composition and timing, then the only way to
answer your question is to get an actual sample of the workload and
determine the composition and timing of the workload and size accordingly.

-- db

----------------------------------------------------------------------
For LINUX-390 subscribe / signoff / archive access instructions,
send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit
http://www.marist.edu/htbin/wlvindex?LINUX-390

Reply via email to