Tom, I agree the block paging technique is a good one, but I wasn't aware block paging was an asynchronous process. I can see where it would be faster when the size or the duration of the sort meant that the block page-in was a cache hit on the storage system. Where the block page-outs are filling cache and being destaged to disk I'm not as certain it will be faster than CFW IO to SORTWK.
Page dataset writes will use more cache on HDS, and fill NVS on IBM much faster than CFW to SORTWK because there is only one copy of CFW, and the cache algorithms, on HDS at least, will try to defer destaging CFW. When the sort is large enough that the read back of block pages becomes a cache miss I think it will not match the IO performance of SORTWK IO when it reads back data. Note I'm just comparing IO performance, there may be other factors in your sort that make it better than DFSORT and SYNCSORT. Ron > -----Original Message----- > From: IBM Mainframe Discussion List [mailto:[email protected]] On Behalf Of > Tom Harper > Sent: Wednesday, March 23, 2011 4:32 AM > To: [email protected] > Subject: Re: [IBM-MAIN] Region size > > Although I am not familiar with the internal design of DFSORT and other major > sort products, it is certainly possible to design a sort which uses the page > data sets by explicitely paging blocks of data out and back in without using > demand paging. We designed our own sort which has just that capability and it > runs considerably faster than running with SORTWK data sets. > > This design does have drawbacks, such as the dedication of space for page data > sets, which lasts for much longer than SORTWK data sets. However, with th > price of DASD so low, it scarcely makes any difference any more. > > Tom Harper > IMS Utilities Development Team > Neon Enterprise Software > Sugar Land, TX > > -----Original Message----- > From: IBM Mainframe Discussion List [mailto:[email protected]] On Behalf Of > Ron Hawkins > Sent: Wednesday, March 23, 2011 1:16 AM > To: [email protected] > Subject: Re: Region size > > ROTFLMAO! They thing demand paging will be faster than CFW to SORTWK? Oh the > pain... > > > -----Original Message----- > > From: IBM Mainframe Discussion List [mailto:[email protected]] On > Behalf Of > > Mike Schwab > > Sent: Tuesday, March 22, 2011 1:41 PM > > To: [email protected] > > Subject: Re: [IBM-MAIN] Region size > > > > I have some users who requested 150GB of Paging packs so they could do > > an in core reorg of multiple M9 volume DB2 databases, because that was > > the fastest way. I told them that was too much and they needed to > > specify sort work statements. They resisted and the request is > > pending. > > > > ---------------------------------------------------------------------- > 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
