On Dec 12, 2007 4:18 PM, Huegel, Thomas <[EMAIL PROTECTED]> wrote:

> When I do this calculation I come up with a requirement of 10 3390 mod3 for
> my page space.
> I only have 2 3390 mod3's for page and I don't seem to be having any
> problems.

When you run out of paging space, CP will abend typically faster than
you can say Oops.

The calculation does the worst case scenario when every user wants
what you allowed him to have. Linux is that worst case. CMS users take
what they need, Linux takes what you let it have. And I trust you also
added the VDISKs to your calculation...

If you only have 6 GB of paging space, it does not take that many
users doing something unusual before you fill it up. Because Linux
roughly uses the entire virtual machine size in a LRU fashion, at the
moment it does not fit entirely in real memory, it will end up
entirely on paging DASD. And not just that one Linux machine, but all
of them because CP will take pages away from all of them.

Even if you have not configured your paging DASD optimal, you can
probably stand 1000 pg/s. That means that you fill the remaining 50%
of your paging space in less than 10 minutes. Even if you have
monitoring active on that, you still need to be pretty quick to beat
that.

Obviously, when you already have 25 GB of page space used at less than
50%, you would have a lot more time before bad things happen.

Trust me, you would not be the first to abend because you run out of
paging space. You know your workload and users best to estimate the
chances that this may happen, and you know whether the investment in
paging space is as insurance is justified.

Rob

PS I hear that some folks recommend to give each Linux server one of
more real disks for swap that do not get used. Combined with the myth
that swap space should be twice the virtual machine size, they end up
with an unused disk of twice the virtual machine size. Clearly my
choice would be to give it to CP as paging device...  and let Linux
(not) swap into VDISK.
-- 
Rob van der Heij
Velocity Software, Inc
http://velocitysoftware.com/

Reply via email to