> RSM will free frames if available frames are in short supply.
> If you obtain a smaller amount, get it backed, and then release it, you
> may be more likely to observe the frames being retained.
I've changed the program to allocate 8KiB areas instead of 10MiB, and RSM
decides to keep the
>If you obtain a smaller amount, get it backed, and then release it, you
may be more likely to observe the frames being retained.
Thanks. I (wrongly, it seems) assumed it is more likely to happen with large
areas.
--
Peter Hunkeler
>It avoids the issuance of IPTE (an expensive instruction when many CPs are
>present) for each frame...
I know. Had I not been too lazy to open the PoOp, and read what exactly the
IPTE does, I had written "it avoids updating the TLB..." instead of "it
probably avoids...". I promis to do
On 12/14/2017 1:13 PM, Peter Hunkeler wrote:
The APAR decribes that marking frames as freemained frames avoids signaling all
processors, which does not scale well with the large number of processors in
z13 and newer machines. It probably avoids updating the TLBs.
It avoids the issuance of
>My guess is that the TLB has not been updated yet. That save performance as
the unallocation can wait until all CPs running in the address space have had
normal interrupts to process the changes.
The unallocation can wait until RSM sees a need to put them onto the available
frames queue. It
>This not much different from the old situation, where you getmain a page,
>freemain half of it, so the frame still remains valid and is fully accessible.
Yes, it is much different! In the case you describe, you will find the page is
allocated in the VSMDATA report. With freemained frames,
On 12/14/2017 6:47 AM, Peter Hunkeler wrote:
See OA46291. The APAR is for z/OS V2.1 and V1.13, it is in the base for z/OS
V2.2.
Also this great write-up put together by Elipda Tzortatos
http://publibz.boulder.ibm.com/zoslib/pdf/OA46291.pdf
With this new function, RSM (SRM?) may decide
>What does this exactly mean: "freemained frames"? Is it just a counter or do
>they refer to specific frames, that are kept at hand in the address space for
>faster assignment of new frame requests?
>
>I am not surprised that those frames, if they are "available" at all, are not
>accessible