Martin, still not sure that I get your point :-)
>I'm wondering whether one wouldn't be better off making the structures >bigger and (perhaps) the buffer pools in the z/OS systems smaller. >In other words, what are the trade offs for biasing towards space in the >CF list structures vs biasing towards space in XCF buffers in z/OS? And here's why: I don't think XCF/XES work like this! Take the xcf signalling structures: You size them with the sizer, you get asked for number of systems that will connect and for the classlen which determines the maximum message length such a signalling structure can handle. That determines the number of 'needed' signalling paths for full connectivity. In reality, XCF uses your (init-)size-specification and allocates the structure. Each 'signalling path', i.e. each connection from one system to another, uses one list in that list structure. The (INIT)-SIZE specification basically determines how many lists with a certain entry-to-element ratio can be allocated in the structure. As far as I know, overspecifying the size in the policy basically wasts CF space, as those lists (aka signalling paths) are not used! Lock structures are different in that the hashing algorithm used basically determines where the 'quantum jump' is in termes of how many 'contentions' get the same size structure. I am assuming (Bill?) that the hashing algorithm used somehow influences the cfsizer, which is why ISGLOCK shows different size numbers than IRLM lock structures when inputting the contention information. Now cache structures, let's look at RACF :-) The nature of a cache structure is that it caches data, so I'd say biasing towards CF storage versus MVS storage depends on the application, what it caches, what type of cache is used, what serialization protocol is used for updates. There may be situations where more CF storage offers an advantage over bigger 'local' cache, even when serialization overhead is involved. You would probably know this for DB2 group buffer pools :-) But (you knew there was a but, right?): We're back to signalling structures, and here the parm that influences the 'buffer pool size' owned by XCF is called MAXMSG. There's a section in the sysplex setup book talking about maxmsg, which is the allowed upper limit for the 'pool'. Okay, that storage must be backed DREF as the code actually accessing it runs disabled. Assuming a properly configured system, *if* that buffer fills to capacity (and hence impacting paging and performance), chances are something is wrong with either another system (because it is probably not reading out the appropriate list in the signalling structure, causing it to fill until no one can write into the structure anymore, and XCF has to retry the CF write operation until there's space) or with your own system because *someone* isn't doing the msgin service properly. Wouldn't you rather have the application unable to go on talking to other members in the group (i.e. msgout gets a return code) rather then XCF having to organize huge storage areas and retrying CF operations? And running the danger of loosing not only this, but other systems as well? (I certainly would, but I had looked at sadumps where *something* was wrong :-) ). So in essence, I think the answer is 'it depends'. Biasing towards CF structure size in my opinion won't buy you anything in terms of performance for list structures, at least not for signalling structures. For lock structures you may gain something if you're able the determine the point of the 'quantum jump'. And for cache structures it depends on how they are used. Okay, now over to others..... Best regards, Barbara -- GMX startet ShortView.de. Hier findest Du Leute mit Deinen Interessen! Jetzt dabei sein: http://www.shortview.de/[EMAIL PROTECTED] ---------------------------------------------------------------------- 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