Re: Memory and CPU Considerations for IBM i LPARs

https://www.ibm.com/support/pages/performance-and-capacity


From: MIDRANGE-L <midrange-l-bounces@xxxxxxxxxxxxxxxxxx> on behalf of Jacob Banda via MIDRANGE-L <midrange-l@xxxxxxxxxxxxxxxxxx>
Date: Wednesday, August 21, 2024 at 4:08 PM
To: midrange-l@xxxxxxxxxxxxxxxxxx <midrange-l@xxxxxxxxxxxxxxxxxx>
Cc: jnbanda_hecorp.com <jnbanda@xxxxxxxxxx>
Subject: [EXTERNAL] Memory and CPU Considerations for IBM i LPARs
So I have a new Power 10 set up with IBM i hosting i and Linux partitions,
managed with V10R3 vHMC. I have a good understanding of the workings of
Processing Units (PU) vs. Virtual Processor (vCPU) allotments in the
partition profiles and have been monitoring the workloads with the HMC
Performance Dashboard for Average and Maximum CPU consumption.

My questions are regarding assigning sufficient CPU resources to the client
partitions and identifying memory usage in the IBM i client partitions.

1. How the heck do I identify just how much memory is required for my
workloads in an IBM i partition? DSPSYSSTS just seems to dump the total
system main storage (memory), but I have a feeling that it's allocating
everything because we gave it a lot to begin with. How can I identify how
much I can cut back? I would REALLY like to cut back some memory on
partitions that we migrated from dedicated boxes with only 1 partition per
box.

2. Is there a rule of thumb for the PU to vCPU ratio on Power hardware? On
my x86 hosts I tend to shoot for 1:2. For mostly idle workloads, I use 1:4.

3. For Shared Processor Partitions, do you all try to set Entitled Capacity
(PU) to be close to average usage, and vCPU to peak usage, with the MAX PU
value set to match the vCPU max, or as close as possible? I read this on a
slidedeck published by Bjorn Roden during an IBM Edge 2015 conference.