On Feb 6, 2008 11:34 AM, Ceruti, Gerard G
<[EMAIL PROTECTED]> wrote:

> The remark to Mark was perhaps to quick off the keyboard , as one big
> issue I am having at the moment in motivating the possible migration of
> applications to zSeries Linux is the amount required and then the cost
> of zSeries memory.

Something lost in the comparison with AIX is that we have 2 levels of
sharing involved. The first level is Linux weigh the SAP virtual
memory against its other use for memory (like page cache). This is
similar to what you see on other platforms.
But the next level is z/VM paging the entire virtual machine. That's
something you don't have on LPAR on pSeries.

Clearly z/VM could not actively page major portions of the working set
without the virtual machine notice (i.e. suffer).
But when we deal with low utilization servers, we *can* use paging to
take memory resources away from idle servers and give it to the
servers that start a transaction. This way the servers take turns in
using the memory.
It does introduce some latency (slow start of your transaction), but
latency is the only thing you have to offer in return for savings.
There's only "secret storage expander" inside z/VM would be sharing
memory among virtual machines.

The challenge is to get the latency low enough that it meets your
service levels. You get that low by a well-designed paging subsystem,
making sure the z/VM does not get confused about who needs what, and
by reducing your (apparent) per-server memory requirements. Takes
measurements and active tuning.

That's the techniques I used to run 100 Linux servers with apache on a
P/390 with 128M memory (and find CPU was more limiting than memory).

Rob
--
Rob van der Heij
Velocity Software, Inc
http://velocitysoftware.com/

----------------------------------------------------------------------
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