Don, At a guess you probably have eight channels per LCU, and your volumes are configured on RAID-5 with 6+P+S or 7+P. In terms of potential parallel IO that makes eight locals look really good as a starting point for handling your peaks.
If you have more paths and array groups then you can always have more than eight locals. You can use Custom Sized Volumes to allocate many, small volumes for Locals, or with Dynamic PAV you can put many small locals on one volume - one volume still has eight channels and 7 or eight underlying HDD. Note that the number of HDD is important for Page-ins, as they are usually cache misses. EMC and HDS customers can use Permacache or FlashAccess to create cache-locked Page Datasets for better-than-Flashdrive paging performance. Ron > -----Original Message----- > From: IBM Mainframe Discussion List [mailto:[email protected]] On Behalf Of > Don Williams > Sent: Thursday, January 21, 2010 8:42 AM > To: [email protected] > Subject: Re: [IBM-MAIN] PAGE datasets -- few large or more small > > When all is well with the world, our paging is near zero. With response time > sensitive users (doctors, nurses, etc.), if we are paging, we buy more > memory. So the main purpose of our paging system is just to provide the > required backing for the all the virtual memory in use. However, in the real > world, stuff happens so you have to be able to handle those problems > quickly. When all hell breaks out, our paging can peak to 100 to 200 pages > per second (e.g., during an SVC dump). So the paging system has to be able > to handle that peak. > > Historically, I have followed the 25% to 30% ROT for block paging. However, > since we do so little paging in general, I not really sure how much benefit > we are getting from that, but it cheap to do it. > > My real question was really more pointed to the fact that we have only 1 > DS8000 controller. In other words, we cannot spread the paging across > multiple physical controllers. We can spread things over many virtual > control units and many virtual volumes, but they are all on the same > physical hardware. Since I'm not well versed in the DS8000 architecture, I > have to treat the DS8000 like a black box, therefore it is difficult to know > which configuration would work best. > > Thanks, Don > > ---------------------------------------------------------------------- > 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
