Re: Testing hardware RESERVE
On 25 January 2011 11:18, Anne & Lynn Wheeler wrote: > Another approach ... is a CKD channel program with "compare&swap" > semantics that was developed for HONE in the late 70s (US operation was > possibly largest single-system-image, loosely-coupled operation in the > world at the time) ... was more efficient than RESERVE/RELEASE (but not > as efficient as ACP RPQ) ... since it involved additional rotation. At > one-time there was extensive discussions with the JES2 multi-spool group > doing something similar. JES2 did implement that; they called it their "atomic" channel program. I remember being at the SHARE session where they announced it, but of course they didn't mention HONE or VM. Tony H. -- 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
Re: Testing hardware RESERVE
paulgboul...@aim.com (Paul Gilmartin) writes: > Long ago, circa MVS 3.8 without GRS, in our little lab we got > sporadic deadlocks when one job allocated SYSLIB on VOL001, > SYSLMOD on VOL002, and another allocated SYSLIB on VOL002, > SYSLMOD on VOL001. long ago and far away ... discussion of the ACP RPQ for 3830 ... allowing for fine granualarity locks (more like VAX/VMS) in lieu of reserve/release http://www.garlic.com/~lynn/2008i.html#email800325 in this post http://www.garlic.com/~lynn/2008i.html#39 American Airlines above references System/R which was the original relational/SQL done in bldg. 28 http://www.garlic.com/~lynn/submain.html#systemr Another approach ... is a CKD channel program with "compare&swap" semantics that was developed for HONE in the late 70s (US operation was possibly largest single-system-image, loosely-coupled operation in the world at the time) ... was more efficient than RESERVE/RELEASE (but not as efficient as ACP RPQ) ... since it involved additional rotation. At one-time there was extensive discussions with the JES2 multi-spool group doing something similar. Misc. past posts mentioning internal HONE system http://www.garlic.com/~lynn/subtopic.html#hone Later I needed a inverse of RESERVE for large cluster operation ... in recovery operation needing to remove a specific processor from the configuration ... I wanted a FCS switch operation to allow everybody ... but the processor that has been removed from the configuration (there is a failure mode where a processor stops, appears to have failed, and then later resumes ... potentially just before doing some I/O operation with global impact, aka it doesn't realize that it has been removed from the configuration). One of the problems was that FCS was being quite distracted with the complex effort to layer FICON on top of it (somewhat in the manner that *ALL* current day CKD is done by simulation on top of underlying FBA). -- virtualization experience starting Jan1968, online at home since Mar1970 -- 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
Re: Testing hardware RESERVE (was: no subject)
On Tue, 25 Jan 2011 08:58:29 -0600, Walt Farrell wrote: >On Tue, 25 Jan 2011 01:56:51 +, john gilmore wrote: > >>I suggest that Mr. Rowe conduct a simple experiment, as I just did. Bind a >set of object modules into library >>A and write something into dataset B, located on the same notional volume. > He will find that he can do these >>operations concurrently, which would be impossible if a hardware reserve >were in use. > >Even then, the most that will happen is that the bind operation for A would >delay the write to B until it has finished, so unless you can stop the bind >operation in the middle, while it holds the serialization, I don't see how >you can really prove anything. > Long ago, circa MVS 3.8 without GRS, in our little lab we got sporadic deadlocks when one job allocated SYSLIB on VOL001, SYSLMOD on VOL002, and another allocated SYSLIB on VOL002, SYSLMOD on VOL001. -- gil -- 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
Testing hardware RESERVE (was: no subject)
On Tue, 25 Jan 2011 01:56:51 +, john gilmore wrote: >I suggest that Mr. Rowe conduct a simple experiment, as I just did. Bind a set of object modules into library >A and write something into dataset B, located on the same notional volume. He will find that he can do these >operations concurrently, which would be impossible if a hardware reserve were in use. It's only impossible if you try that from 2 different systems, and if you can really ensure that the operations on those two systems are truly simultaneous. That's very difficult to arrange. Even then, the most that will happen is that the bind operation for A would delay the write to B until it has finished, so unless you can stop the bind operation in the middle, while it holds the serialization, I don't see how you can really prove anything. -- Walt Farrell IBM STSM, z/OS Security Design -- 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