On Tue, 8 Sep 2009 16:42:25 -0500, Mark Zelden <[email protected]> wrote:
>On Tue, 8 Sep 2009 17:11:25 -0400, Joel Wolpert <[email protected]> wrote: > >>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 > >Good point. I assume your statement is related to DB2 CPU consumption >only, not the entire system. > >I tend to live in my own little operating system world, so >my numbers were just related to OS overhead. I was part of many >parallel sysplex implementations in the past, but never have been >at a shop "just before" and "just after" a DB2 datasharing implementation >to compare numbers. I haven't looked at this aspect in a while, but I >think the DB2 manuals may quote a similar number (10%). > >As it turns out, I think we will be implementing DB2 data sharing here >for one of our environments that doesn't utilize it now. The access >between 2 LPARs is currently DDF (formerly type 2 connector, now >type 4 / DRDA to utilize zIIP) - so overall we expect this to be an >improvement. When it was type 2, it was more clear on what the MIPS >and dollar savings would be, but with zIIP thrown in the picture now, >I'm not sure. Regardless, performance will be better for the LPAR that >is not local today but the current DB2 "owning" LPAR will take that hit >that you are referring to. > I couldn't find anything in black and white about this, but did find the following in an old data sharing implementation Redbook - DB2 for MVS/ESA Version 4 Data Sharing Implementation (SG24-4791-00) that just states what has already been said - ymmv: >From Chapter 10. Performance and Tuning Considerations: "It is almost impossible to determine in advance the overhead caused by data sharing in your environment because this overhead depends mostly on your workload profile, degree of data sharing, physical database design, bind options, and hardware configuration." In the Redbook - DB2 for z/OS:Data Sharing in a Nutshell (SG24-7322-00), chapter 7.2 on performance also states it is workload dependent for a 2-way group but it does document an actual number for each additional member - "typically <1% overhead". According to a DB2 sysprog who was here when DB2 data sharing was implemented in one of our large sysplexes, our overhead was about 8-9%. DB2 uses about 50% of the cycles so that is (was) a significant number. Does anyone else have any real world numbers they can share of what their overhead was when they implemented DB2 data sharing (I supposed this sort of information could be in a DB2 listserv archives)? 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
