Interesting patent related to SPKA

http://www.google.com/patents/EP0558945A2?cl=en

Rob Schramm
Senior Systems Consultant
Imperium Group




On Thu, Nov 15, 2012 at 4:17 PM, Schwartz, Alan <[email protected]>wrote:

> There's a conflict here.  The CA APAR says  " This APAR will have a
> greater affect on regions running STGPROT=NO."
> Jim said " Therefore, on workloads (i.e. CICS running with
> STGPROT=YES).... , this can show up as the z196 spending more time"
>
> Can't be both.
>
> Alan Schwartz
> ITO Global Services Operations and Engineering
> Xerox Business Services, LLC
> 1500 Towerview Rd.
> Eagan, MN. 55121-1346
>
> p.  612.266.3150
> m. 651.274.5819
> f.   612.266.3196
>
>
> -----Original Message-----
> From: IBM Mainframe Discussion List [mailto:[email protected]] On
> Behalf Of Jim Mulder
> Sent: Thursday, November 15, 2012 12:57 PM
> To: [email protected]
> Subject: Re: CA Common services ENF Monitor reporting high CPU time
>
> > We observed, while using STROBE, apparent high CPU use in module
> CAS9C66.
> > From CA we found that they had the following on file::
> >
> > APAR #: RO43562
> > Product: ENFCIC     Release: 14.0        Solution #: 7         Type:
> >      OS: OS           Group: GCCOMC ISL SUP 2
> >   Title: PERFORMANCE ENHANCEMENT ON Z/196 PROCESSORS.
> >
> > ****************** VERSION 0   EFFECTIVE: MAR 31 2012   2:09
> > ******************
> >
> > ***NOTE***  PE: YES   CORRECTED BY:  RO45646
> >
> > PROBLEM DESCRIPTION:
> > After a processor upgrade to z/196, some performance monitors may show
> > increase cpu usage in various csects in the CAS9Cxx module.
> > The flagged area is usually in a very tight range and will contain a
> > SPKA instruction. This APAR will have a greater affect on regions
> > running STGPROT=NO.
> >
> > SYMPTOMS:
> > Performance monitors show increased activity in CAS9Cxx modules.
> >
> > We had just moved to a z196 and STROBE was being used to compare
> > performance against the previous processor (no longer available for
> > direct comparison).
> >
> > Since the z196 and z114 are from the same design cycle and zEC12 is
> > similar to a z196 we have asked whether this situation could exist on
> > a z114 or a zEC12 and they said no.
> >
> > Has anyone any idea why this might occur, and anyone has seen it on a
> > z114 or zEC12.
> > Is the set up code for a SPKA instruction something that would be very
> > different on a z196 from any other processor?
>
> One of the CPU designers gave me the following explanation:
>
> System z processor development has identified an aspect of the z196
> processor that performs worse than the equivalent instruction on a z10
> processor. When an SPKA instruction is executed in problem state, the new
> out-of-order design of the
> z196 processor requires more pipeline stalls to give functionally correct
> results than in prior generations of processors.
>
> Therefore, on workloads (i.e. CICS running with STGPROT=YES) that have an
> intense amount of SPKAs in problem state, this can show up as the z196
> spending more time executing the SPKA instruction. Some vendor performance
> tools or single instruction benchmarks may uncover this additional time
> spent on the SPKA instruction. This change in SPKA behavior does not offset
> the benefits the z196 provides for the CICS environment.
> This aspect of the longer SPKA execution time can be exacerbated by
> running on a subcapacity machine.
>
>
> Jim Mulder   z/OS System Test   IBM Corp.  Poughkeepsie,  NY
>
> ----------------------------------------------------------------------
> For IBM-MAIN subscribe / signoff / archive access instructions, send email
> to [email protected] with the message: INFO IBM-MAIN
>
> ----------------------------------------------------------------------
> For IBM-MAIN subscribe / signoff / archive access instructions,
> send email to [email protected] with the message: INFO IBM-MAIN
>

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN

Reply via email to