John, I'm relying on my memory and not an archive search. I'm sure there is a post from Greg Dyck many years ago saying that paging is optimised for a minimum of eight page datasets.
He was disagreeing with a statement (by me I think) along the lines that fewer, full-pack datasets are just as good as many smaller ones. Every MF RAID vendor support Custom Volume Sizes, so there is no need to have any wasted space whether all eight locals, PLPA and Common are on dedicated volumes, or you simply plonk them all on one or two volumes with a dedicated LCU and enough Alias assigned. If I am right that resume subchannel is no longer used by ASM, than putting other low to medium use datasets on the same volume will make little or no difference to paging performance. Ron > -----Original Message----- > From: IBM Mainframe Discussion List [mailto:[EMAIL PROTECTED] On > Behalf Of John Laubenheimer > Sent: Wednesday, May 07, 2008 2:38 PM > To: [email protected] > Subject: Re: [IBM-MAIN] 3 Page Datasets on one Volume > > On Wed, 7 May 2008 14:58:21 -0500, Staller, Allan > <[EMAIL PROTECTED]> > wrote: > > >It depends. > > > >In the old SLED days this could be performance crippling, especially > if > >there was a decent paging rate(remember only one actuator/volume). > With > >little or no paging this would not be a large problem. > > > >Fast forward 20 years: > > > >Data is mapped transparently to many many small drives, accessed by > many > >actuators. Again if there is no significant paging, no problem. If > >significant paging occurs, this MAY BE a problem, maybe not. > > > >Having said all of the above, my distinct preference would be for one > >page ds the size of the 3 currently occupying the volume. > > > >HTH, > > > > > ><snip> > >I noticed again today that we have 6 local page datasets on 2 volumes > - > >each a 3390-3. I'm sure they have Pavs, or the equivelent on the > dasd, > >but I'm just wondering if that is a good practice or not. > ></snip> > > > > > >---------------------------------------------------------------------- > >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 > It is best to have a minimum of four(4) local page datasets; more is > better. > The heavy hitter that gets lost in the mix is an SVC dump. This can > put quite > a load on the paging configuration when it needs to page-in a lot of > inactive > frames, just to write them out to a dump dataset. The system will be > disabled > while the copy process copies this to a dataspace (which, in turn, can > stress > the real storage configuration, and thus cause more page-out activity). > You > really want to be disabled for as short a time as possible. So, more > concurrent > paging I/Os is better; the way to get that is more local datasets. > With PAVs, > they need not be on different volumes. Without PAVs, separate volumes > are > strongly recommended. > > ---------------------------------------------------------------------- > 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
