Unsurprisingly, it looks as if I need to clarify a little ...

Tom's not happy with, "assuming" R/R support for Reloc 0 Minidisks. My 

take here is that CP is not unilaterally adding any R/R activity - if the
 
disk at Reloc 0 is a VSE minidisk and VSE doesn't issue any R/R then CP 

won't either - but if VSE DOES, "feel the need" then the R/R will be 
honoured by CP and protection from other systems running in other images 

will be provided - just as it would be in the, "Real World". This being 

the case, I don't see how with R/R propagation at the, "real" level we're
 
significantly worse off than we are today - if VSE doesn't do it then CP 

doesn't do it - if VSE DOES do it then it gets the protection it's asking
 
for instead of a weak subset of what it's expecting. There IS a new impac
t 
(in terms of responsiveness) for users of other minidisks that share the 

real volume - but I would far rather manage the impact of coping with the
 
consequences of, "safe" behaviour than manage the impact of coping with 

the consequences of a data corruption caused by CP's failure to fully 
honour a request from a guest asking for short-term exclusive use of a 

minidisk.

(I contend that the CPU overhead of checking for the need to drive R/R 

support is not worthy of consideration - we are talking a few tens of 
instructions increase in the normal pathlength at most and - to my mind -
 
the benefit of consistent behaviour far outweighs the cost of these few 

instructions.)

Richard's appears to be unhappy with me suggesting that XLINK has a part 

to play here. I think that Rob's response should have settled these 
concerns but, just to make sure, a reminder that XLINK extends the 
protection against, "unintentional" multi-write access that one gets on a
 
single system to cover multiple systems sharing a single real volume. It 

does this via a, "Compare-and-Swap" mechanism on a cylinder-map that 
resides on the real volume and it requires no additional hardware or (onc
e 
set up) support process. Again, as it's intrinsically dangerous to have 

multiple CP images sharing minidisks on a real volume without the 
protection of XLINK, and XLINK is part of the base system, why on Earth 

would anyone wish to operate a production system without it? That being 

the case, it seems natural to me to deliver real R/R in support of any 

virtual R/R issued by a guest to any minidisk with a, "V" in the MDISK 

statement (or starts at Cylinder 0) that's on a real volume under the 
control of XLINK. In the vast majority of cases the non-reloc-0 minidisks
 
will be CMS minidisks and (because CMS doesn't use R/R) there will not be
 
any reason to code the, "V" anyway - but having the support in CP gives u
s 
the CAPABILITY (which we don't currently have) should we wish to make use
 
of it for - say - a VSE LOCKFILE residence minidisk.

(At least nobody's yet taking me to task for suggesting that volumes unde
r 
XLINK control should default to SHARED when ATTACHED to SYSTEM! <grin>)

Hopefully I've been able to make my points somewhat more clearly this tim
e 
around. Tom and Richard - through your comments, thanks for giving me the
 
opportunity to do so.

Regards
Jeff

Reply via email to