Coincidentally we had Earl working with us earlier this week and I read your question to him - he said go look at the IBM Virtual User Group presentation from October 6, 2022.
My question (from spending time with Earl and hearing this point) - are you running over entitlement? That topic as regards ilcs is covered in that presentation.
Original Message:
Sent: Mon October 02, 2023 09:57 AM
From: Henrik Morsing
Subject: ilcs on only LPAR in pool
Thanks Jakub,
That is very interesting!
Regards,
Henrik Morsing
------------------------------
Henrik Morsing
Original Message:
Sent: Mon October 02, 2023 09:36 AM
From: Jakub Pacowski
Subject: ilcs on only LPAR in pool
From: https://www.redbooks.ibm.com/redbooks/pdfs/sg248535.pdf page 83
"Levels of processor capacity resolution
Two levels of processor capacity resolution are implemented by the PHYP and multiple
shared processor pools (MSPP):
Level0 The first level, Level0 , is the resolution of capacity within the same
SPP. Unused processor cycles from within an SPP are harvested and
then redistributed to any eligible micro-partition within the same SPP.
Level1 When all Level 0 capacities are resolved within the MSPP, the
hypervisor harvests unused processor cycles and redistributes them
to eligible micro-partitions regardless of the MSPP structure. Level 1 is
the second level of processor capacity resolution"
LPAR has to cede unused cycles (vlcs) to allow other partitions to use it, if it is below desired processing unit limit (guaranteed CPU).
Above desired processing unit value lpars will compete for the cycles and hypervisor can "steal" the cpu for other partitions (ilcs).
Max pool size is just limit of cpu in the pool. Nothing more.
------------------------------
Jakub Pacowski
Original Message:
Sent: Mon October 02, 2023 09:05 AM
From: Henrik Morsing
Subject: ilcs on only LPAR in pool
Hi Jakub,
Are you saying another pool can consume cycles from this pool, and kick off an LPAR in this pool? I don't think that's right.
Also, the LPAR here is not even close to hitting the limit in its pool.
Regard,
Henrik Morsing
------------------------------
Henrik Morsing
Original Message:
Sent: Mon October 02, 2023 08:36 AM
From: Jakub Pacowski
Subject: ilcs on only LPAR in pool
Henrik,
if you have single LPAR in the pool then this partition can take free cpu cycles only from reserved capacity (if set up) or from partitions outside the pool (sharing CPU is done on two levels -> inside a pool, and global to the server). Imho it's perfectly normal that your partition can use all of it's cycles and hypervisor need to schedule them to other lpars. In case of high load it will cause visible ilcs stats. Max pool size is just limit. It does not guarantee you that your lpar will get any cycles.
vlcs - voluntary logical processor context switches - means partition has nothing useful to do and cedes cycles to hypervisor. In that case hard to say about performance impact. The only risk I can see is "bursting workload" when you have periods of high load and periods of idle system. LPAR cedes cycles to cpu but when it needs them to cover high load period it will get them with delay. I would say it's not typical type of workload.
------------------------------
Jakub Pacowski
Original Message:
Sent: Mon October 02, 2023 08:13 AM
From: nigel griffiths
Subject: ilcs on only LPAR in pool
Can you give us a few clues:
What OS is the LPAR running?
Which tool are you getting ilcs from?
What Hardware? P7 P8 P9 P10
That Model? Like S1024
Which OS(s) are you using?
Sent from Mail for Windows
Original Message:
Sent: 9/29/2023 3:33:00 AM
From: Henrik Morsing
Subject: ilcs on only LPAR in pool
Good morning,
Does anyone know ilcs can be non-zero on a single LPAR in a pool? It never gets close to maxing out the pool.
Also, how harmful is high vlcs numbers to performance, I can't specifically find anyone saying it's bad.
Regards,
Henrik Morsing
------------------------------
Henrik Morsing
------------------------------