Tom Schmidt wrote:
On Mon, 19 Dec 2005 09:56:37 +0100, R.S. wrote:
Tom Schmidt wrote:
On Sun, 18 Dec 2005 14:54:15 +0100, R.S. wrote:
John S. Giltner, Jr. wrote:
[...]
Mostly Linux on the MF was so that you could migrate what was on 10, 20,
200 Intel boxes that were sitting there 90% idle and put on a single box
and share CPU and Memeory resources.
IMHO this is marketing mantra from IBM.
1. Sharing memory resource is not true. Memory can be easily
reconfigured (taken from system A to system B), but, assuming 24x7
availability it is big advantage.
1a) Sharing memory resource IS true if you are using z/VM, as the site
where I'm backup VM-guy is doing with its mainframe Linux workloads.
They have 24x7 (well, closer to 24x6.98 with periodic scheduled outages
for non-VM reasons). If it was important to the business to be fully up
at 5am on every Sunday morning they could achieve 24x7 but it is more
important to the business that it spend that money elsewhere.
Does it mean, the system takes memory from common pool when needed ?
Or you can configure some memory offline from one system and assign it
to another, everything without any IPL of any system (mean Linux image) ?
My humble opinion is that you can reconfigure memory between Linux
virtual machines but only when these machines are down. Maybe I'm wrong
with that.
You're wrong with that. VM's virtual memory implementation shares among
all of its Linux parasites ALL of the time. The LPAR memory is owned by VM
and a subset of that is available to the pool of Linux systems. The size
of the subset varies, based on many factors but it is generally a healthy
number - plenty for our purposes.
One thing for you to consider, R.S., is that Poland is approximately the
size of Wisconsin. Our usage extends beyond our borders - worldwide,
actually. Because of the reach across many timezones we are more likely to
see substantial variability in usage from Linux instance to Linux
instance. If you only work within one timezone you might not find the same
opportunities.
Yes, I'm aware of timezones. I considered to mention it. However I doubt
if different timezones are served by different servers. I would vote for
the opposite. For example, we run two banks on one CPC, different LPARs.
Unfortunately, the traffic comes from one timezone (although we're open
24x7).
Ab ovo: all the machines, which constitute information system, working
together, have the same rush hours. Exception for that could be backup
servers, batch machines, etc. Usually CPU power is sized to fulfill
on-line requirements, not batch or backups duration. Especially, if we
talk about application servers, webservers, etc.
BTW: Thank to Jon Brock who confirmed my doubts. It seems I can increase
size of virtual machine's memory, but Linux image won't use it until
re-IPLed.
Disclaimer: I'm not neither mainframe nor z/Linux enemy. I just try to
discuss real costs or savings from using IFL. Sometimes hardware RAS can
justify it, sometimes not, especially when Linux images are used for non
mission-critical applications. Nowadays PC servers can be really
dens-packed, can use the same (or much cheaper) SAN storage, can be
well-administered en-mass. 100 Linux images mean 100 operating systems,
100 root users. On 100 PCs or single VM, doesn't matter.
--
Radoslaw Skorupka
Lodz, Poland
----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO
Search the archives at http://bama.ua.edu/archives/ibm-main.html
