Guidelines for Configuring Virtual Partitions on Cellular Platforms

16
5. CLM and CLPs. vPars customers are encouraged to assign CLM and CLPs for each of
their virtual partitions
6
. It is better to configure some amount of base CLM from a cell if the
virtual partition is going to have processors and I/O on that cell, as this will allow the kernel
to allocate the I/O related data structures in the same locality. When configuring CLM and
CLP for each virtual partition, refer to the HP-UX NUMA documentation to decide on how to
optimize the performance [1,2].
6. Multi-core CPUs. Some multi-core CPUs have a second or a third level cache that is
shared by the cores on that socket. Assigning CPU cores from the same socket to different
virtual partitions may, depending on the nature of the memory usage patterns of the hosted
workloads, lead to some degree of indirect performance interference. Users are encouraged
to do some baseline studies and if needed try to avoid or minimize CPU core assignments to
different virtual partitions from a CPU with shared caches.
7. Base versus floating memory. There is a minimum amount of base memory that
needs to be configured in a virtual partition to allow the Operating Environment to boot and
run. Refer to [4] for more details. Users are encouraged to do some baseline performance
studies for their workloads and then tune the amount of base and floating CLM and ILM
memory in the virtual partition.
8. Resource migrations. While performing dynamic resource migration operations like
adding/deleting memory or adding/removing CPU cores from a virtual partition there is a
need to take into consideration the resource localities. Assigning processors from one cell
and cell local memory from another cell might lead to less optimal performance due to
increased distance between processor and memory locality. Hence addition/deletion of
CLM should also take into consideration the CLP assignments and vice versa.
9. I/O Interrupts. The kernel assigns I/O interrupts during boot time amongst the active
CPU cores in an OS instance using a simple round robin algorithm. (In HP-UX 11i v3, the
I/O interrupt assignment takes locality into account when assigning interrupts at boot time).
An administrator can see these I/O interrupt assignments and change them using the intctl
command. During CPU core deletion any assigned IO interrupts are reassigned to a different
active CPU core in the OS instance. Following a CPU core addition, the new active CPU
core is not automatically assigned I/O interrupts. Repeated CPU core migrations (deletions
followed by additions) can lead to an imbalance in the I/O interrupt assignments amongst
the active CPU cores in the OS instance. I/O intensive workloads can experience
performance penalties due to this. Users are encouraged to follow the guidelines mentioned
in the white paper on CPU Configuration Guidelines [3] to avoid this problem.
6
When HP Integrity Essentials Global Workload Manager (gWLM) is configured/initialized in a vPars environment it discovers the CLP’s and
CPU cores specified by hardware paths and only decides to operate on the CPU cores specified by count and the remaining un-assigned CPU
cores. Any changes to the vPars CLP’s or the CPU cores specified by hardware paths will require an explicit update of the gWLM
configuration. Please refer to the gWLM documentation [10] for further details on this.