If I recall correctly there is a big hit (10-20% overhead) for the first 2
systems in a sysplex when doing DB2 datasharing. After that the overhead
drops drastically.
For other sysplex implementations the overhead can be much less.
Joel Wolpert
Performance and Capacity Planning consultant
WEBSITE: www.perfconsultant.com
----- Original Message -----
From: "Mark Zelden" <[email protected]>
Newsgroups: bit.listserv.ibm-main
To: <[email protected]>
Sent: Tuesday, September 08, 2009 2:10 PM
Subject: Re: Sysplex Basic Question
On Tue, 8 Sep 2009 12:01:08 -0500, Eric Bielefeld <[email protected]>
wrote:
I'm curious. If you have a sysplex with say 2 separate CPUs, can you set
that up so that if one CPU crashes, that the other can take over
immediately? I seem to remember reading about that long ago, where you
can
have the same CICSs and DB2 regions all up on both boxes, but all
transactions are routed to one box or the other. Then, if one CPU would
die, the other could take over immediately, with no wait. Is that the way
it works, or am I not remembering correctly? If that is the way it works,
that would mean a lot to that other thread on EMC and PAVs.
I have experience at 2 of my contract jobs in a sysplex environment. The
first job actually had 2 z/900s in the same room. My last job was 2 Lpars
in a sysplex on a z/800. Both were done to save on software costs. As a
matter of fact, on the 1st job, after the holiday season ended they hard
capped both CPUs to I think 85% of capacity, which I'm sure saved them a
lot
of money. They ran CICS on the 5 engine box, and DB2 batch jobs on the 3
engine box. I don't think anyone complained when the cap was put on. I
think the batch spent a lot more time waiting in the input que, but
overall
CICS response time wasn't affected as that processor rarely ran at 100%,
and
when it did the batch just took longer.
I know at my last job with the single z/800, they were told that the
sysplex
overhead was 10-20%. To me, that is a LOT of overhead, but it did enable
them to save money on software. I think they got rid of the system that
needed CICS, so I think by now they probably eliminated the sysplex and
get
a much lower overall cost. I know they also got a z10, so software costs
also were reduced by the z10.
Basic question, but not a basic answer. So I'm not going to attempt to
really answer the first part (I'll let others do that), but in theory, a
properly
configured parallel sysplex can give your applications continuous 24 x 7
availability. Does that mean that someone right in the middle of a
transaction
when a system / subsystem crashes won't ever get "kicked out" or have to
re-enter that transaction? Certainly not. But for a planned outage
there
hopefully should be no hiccups.
I do want to address the 2nd part because 10-20% is an over statement.
From what I recall from various benchmarks and publications, the overhead
to implement a sysplex is about 5% for a system to join the sysplex (that
would be to each of the first 2 systems if there were only 2). Then an
additional 1-2% for each system being added to the sysplex. In other
words, the non-sysplex system being added would incur about 5% additional
overhead and all the other systems would incur 1-2% additional overhead
when the new system was added.
These numbers are over 10 years old at this point, but I don't know if
there
have been similar studies using a more current version of the operating
system and current hardware.
As usual, YMMV.
Mark
--
Mark Zelden
Sr. Software and Systems Architect - z/OS Team Lead
Zurich North America / Farmers Insurance Group - ZFUS G-ITO
mailto:[email protected]
z/OS Systems Programming expert at
http://expertanswercenter.techtarget.com/
Mark's MVS Utilities: http://home.flash.net/~mzelden/mvsutil.html
----------------------------------------------------------------------
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
----------------------------------------------------------------------
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
