<SNIP> > <snip> > Our CPC (2094-S18-708) now has 8 Physical CP and has 1 production LPAR, > 2 maintenance LPARs, 2 ICF LPARs and 2 test LPARs running at the moment. > Currently, all LPARs are equipped with shared CPs and have weight values > assigned. The details are as follows. > > production LPAR 1 > initial LCP 8 > reserved LCP 10 > weight 40 > </snip> > > Are you saying that all LPARs have 8 active(shared)/10 reserved LP's > assigned? > > If so, this is a bad thing to do! PR/SM overhead is excessive (only the > active LP's affect the overhead).
Can you define "excessive" or simply tell us how big overhead would you expect in such scenario? <</SNIP>> I have seen about 15% of capacity in a large (before corrected) scenario. YMMV. <SNIP> > The logical/physical ratio should (ROT) not exceed 2 to 1. > I have pushed this to 3 to 1 with some very low activity test LPARs > where performance was not critical. Let's assume you have 15 LPARs. How would you assign CPs to those LPARs? Is it possible to have CP shared between, let's say, 2 LPARs, but not other 13? As far as I know the only choice is SHARED or DEDICATED. Shared by *all* LPARs. There is also CF CPU(n),OFFLINE... </SNIP> IIRC, the meaning of DEDICATED is that work on this group of CPs will only be dispatched on these CPs. SHARED CPs *may* be dispatched on "DEDICATED processors if resource is available (I am sure someone will correct me this is not true). This reduces context switch between the processors and reduces overhead. CF CPU(N),OFFLINE will reduce the number of logical processors to be dispatched and improve the LP:CP ratio. <SNIP> [...] > No. IBM does not prevent you from defining more LP's than CP's. However > the overhead becomes excessive when the LP/CP ratio exceeds (2 to 3) to > 1. IBM allows to create 60 LPARs on machines with theoretical limitation 32 CPs. z/990D32. </SNIP> See above. IBM does not prevent you from doing this. However, the results may not be something desirable. Kathy Walsh and Cheryl Watson have both consistently recommended that the LP to CP ration be less than 2:1. My experience corresponds with their recommendations. ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
