From: Holger.VanKoll@SWISSCOM.COM
Date: Tue Dec 02 2003 - 05:57:36 EST
I would use wlm only if necessary - additional complexity and
performance-problems are harder to debug.
If you need it to regulate only cpu, however, its fine.
And I would never mix batch-oriented (regarding io) and
small-responsetime oriented programs (short queries from users).
A bigger partition has advantages - if one application doesnt need "its"
cpu, the other can use it. same for filecaching.
But especially for filecaching batch-oriented applications will push ot
the file-cache of interactive ones.
I had this once and had to split the servers - as it was 4.3.3 and you
have no way of controlling filecache there (mount -dio etc.)
Also, bigger partitions might continue to run if a cpu goes bad - a
one-cpu partition will stop.
And you save io-adapters in most cases - network, san. For 4 partitions,
you need 4x2 san-adapters - for 1 big partitions only 2 or 4. Same for
internal disks - 2 disks for each partition.
Small partitions have the advantage that running software doesnt hurt
each other.
Its too hard to say what will be better.
Anyway, I would never buy a p690 to create small 1-cpu partitions -
thats simply more than double as much $ as to buy p630 or p650.
-----Original Message-----
From: Green, Simon [mailto:Simon.Green@EU.ALTRIA.COM]
Sent: Monday, December 01, 2003 6:18 PM
To: aix-l@Princeton.EDU
Subject: Strategies for effective partitioning of p690.
At the moment, we're running two large SP2 clusters at two sites. These
comprise a wide mix of SP2 PCI nodes and SP attached servers.
Our chosen strategy for new servers is to use p690's and LPAR.
One of the next things we're going to be doing is shifting applications
off
old silver nodes - probably onto Regatta partitions. The main reason
for
doing this is that we want the existing applications on those servers to
use
the 64-bit kernel, which is not possible on the silvers. We'll re-use
the
silvers to replace our few remaining MCA servers, and various small
applications and test systems.
The problem, though, is that even a small LPAR will be very powerful for
anything which will run adequately on a silver node at the moment. And
there are some within our company who are reluctant to have small LPARs
in
the first place.
It seems likely that we'll have to have some LPARs with multiple
work-loads
in them - probably with WLM to keep things under control. But should we
be
going for small LPARs so that we can minimise these shared workloads, or
have larger LPARs with many different workloads? Oh: and the different
workloads could well be for different customers, which has the potential
to
make agreeing maintenance outages a bit of a pain.
Fewer LPARs would be easier to manage - much less hassle allocating I/O
slots - but it won't be too difficult to cope with smaller partitions if
we
maintain good records - will it?
Any of you folks have any opinions/experiences? Should we be going for
big
LPARs each with lots of workloads, or lots of small LPARs with as few
shared
workloads as possible? Is there an optimum number of LPARS in a p690?
Simon Green
Altria ITSC Europe Ltd
AIX-L Archive at https://new-lists.princeton.edu/listserv/aix-l.html
AIX FAQ at http://www.faqs.org/faqs/aix-faq/
N.B. Unsolicited email from vendors will not be appreciated.
Please post all follow-ups to the list.
This archive was generated by hypermail 2.1.7 : Wed Apr 09 2008 - 22:17:23 EDT