Dear Christian
In the past 3 years, I investigated 2 cases of IBM i hosting IBM i with a result of consistently non-optimal disk response time for all client LPARs. But the disk response time of the host LPAR is consistently good at all time (it did not run any user workload). Both cases were from S914 machine with all SSD units attached to the base dual-controller FC#EJ1D with 1.8GB write cache.
I did a Google search for any best practice for such an implementation but found none. Do you, by any chance, have such best practice guideline for this?
Here is PDI chart of SSD response time in the host LPAR (never exceed 0.6 msec which is very good):
Here is PDI chart of vdisk response time in the client LPAR (vary between 2-26 msec which is very disappointing):
All PDI wait charts showed small amount of any wait components. So, I see that the bad vdisk response time is the only issue here.
From my observation in trying to identify the cause of this issue, I noticed only that too few vdisk units were created for client LPARs (as IBM i like sufficient number if disk units). But I wonder if there is any other configuration guideline but could not find any yet.
------------------------------
Satid Singkorapoom
------------------------------
Original Message:
Sent: Mon May 09, 2022 03:04 AM
From: Christian Jorgensen
Subject: Power System for IBM i using NVME
Hi Angela.
Have you considered having IBM i hosting the 10 IBM i LPAR's? This would allow the NVMe's to be owned and managed by the hosting IBM i partition - and no VIOS will be involved. And only IBM i knowledge is required to configure this.
Have a look at this YouTube video for more information about the different hosting options and why you may want to have IBM i hosting IBM i:
Selecting you Power System hosting environment
IBM i hosting IBM i works very well in my experience...
Best regards,
Christian
------------------------------
Christian Jorgensen
IT System Administrator
Network of Music Partners A/S
------------------------------