Technical Blog Post
Abstract
Power Linux Implementations based on several items
Body
Blog: Ron Gordon, Director of Power Systems, Mainline Information Systems
Title: Linux Implementation Options for Power Systems
I have been getting questions on the various Linux implementation options on Power Systems. The implementations options basically involve installation of a bare metal system, a PowerKVM virtualized system, and PowerVM virtualized systems.
- Power Systems Model and Hypervisors:
Let me start with a little information on the various Power Systems servers and their uniqueness’s. The LC models only come with OPAL (Open Power Abstraction Layer) which is a light-weight hardware interface for operating systems and virtualization systems. The L model come with both OPAL and pHyp where pHyp is the hardware interface used by PowerVM for virtualization or to do DLPAR implementations without PowerVM. On the L models you can run either OPAL or pHyp but not both at the same time. Via the ASMI (advanced system management interface) you can switch back and forth between these hypervisors at a re-boot of the system. On the Scale-out and Enterprise class Power Systems that do not have the L or LC designation, only pHyp is provided.
Let me start with a little information on the various Power Systems servers and their uniqueness’s. The LC models only come with OPAL (Open Power Abstraction Layer) which is a light-weight hardware interface for operating systems and virtualization systems. The L model come with both OPAL and pHyp where pHyp is the hardware interface used by PowerVM for virtualization or to do DLPAR implementations without PowerVM. On the L models you can run either OPAL or pHyp but not both at the same time. Via the ASMI (advanced system management interface) you can switch back and forth between these hypervisors at a re-boot of the system. On the Scale-out and Enterprise class Power Systems that do not have the L or LC designation, only pHyp is provided.
- OPAL Bare Metal and PowerKVM considerations:
With OPAL you can run Big or Little Endian distributions (Ubuntu, SLES, Redhat) virtualized on PowerKVM or RHEV or Ubuntu's KVM. OPAL is not supported by PowerVM so PowerVM will not run the LC servers. If you wish to run bare metal for maximum performance reasons, you can only run Little Endian versions of SLES, Redhat, or Ubuntu Ubuntu is only Little Endian so no problem. SLES 12 SP1 supports bare-metal on L models - a future update is expected to support the LC models. Since OPAL is only available on the LC and L models, this means that you cannot run bare metal nor PowerKVM on the Scale-out or Enterprise systems.
One might wish to run bare metal in a minimally virtualized environment using DPARs but DPLARs are implemented only in pHyp so this cannot be accomplished on the LC models. On the other hand it could be accomplished on the L models or the standard Scale-out models in conjunction with AIX and IBM i. To me it seems illogical to run DLPAR Linux partitions on the Enterprise class systems.
PowerKVM can be implemented on either the L or LC models but not on the Scale-out nor Enterprise models since they do not have OPAL. PowerKVM comes with its own Linux kernel and can virtualize both BE and LE guests.
With OPAL you can run Big or Little Endian distributions (Ubuntu, SLES, Redhat) virtualized on PowerKVM or RHEV or Ubuntu's KVM. OPAL is not supported by PowerVM so PowerVM will not run the LC servers. If you wish to run bare metal for maximum performance reasons, you can only run Little Endian versions of SLES, Redhat, or Ubuntu Ubuntu is only Little Endian so no problem. SLES 12 SP1 supports bare-metal on L models - a future update is expected to support the LC models. Since OPAL is only available on the LC and L models, this means that you cannot run bare metal nor PowerKVM on the Scale-out or Enterprise systems.
One might wish to run bare metal in a minimally virtualized environment using DPARs but DPLARs are implemented only in pHyp so this cannot be accomplished on the LC models. On the other hand it could be accomplished on the L models or the standard Scale-out models in conjunction with AIX and IBM i. To me it seems illogical to run DLPAR Linux partitions on the Enterprise class systems.
PowerKVM can be implemented on either the L or LC models but not on the Scale-out nor Enterprise models since they do not have OPAL. PowerKVM comes with its own Linux kernel and can virtualize both BE and LE guests.
- pHyp and PowerVM Considerations:
PowerVM will require pHyp so PowerVM cannot be implemented on the LC models. For virtual partitions on the L, Scale-out or Enterprise systems, LE or BE distributions can be the guest, and if optimal performance is required, then Dedicated Partitions should be implemented via pHyp without the need for PowerVM.
PowerVM will require pHyp so PowerVM cannot be implemented on the LC models. For virtual partitions on the L, Scale-out or Enterprise systems, LE or BE distributions can be the guest, and if optimal performance is required, then Dedicated Partitions should be implemented via pHyp without the need for PowerVM.
- Choosing the optimal runtime environment:
I think the process for determining what options for Linux servers should be made are based on the application needs. However, sometimes you already have the “server” and then are limited by what options exist. For example, if you have Scale-out or Enterprise POWER8 systems you can only virtualize with PowerVM. If you have L models you can virtualize both LE and BE distributions with PowerVM. DPLARs, PowerKVM, RHEV, or bare metal depends on which hypervisor you are running. If you have LC models you can run LE only distributions on bare metal or if you are virtualizing withPowerKVM then you can run BE or LE distributions.
If you are trying to determine which server best matches your application need you should start with the application..BE or LE?..then do you need optimum performance….then do you wish a virtualized environment. That should lead you to the correct server selection.
One final thought is that Little Endian distribution can only run on POWER8 systems.
I think the process for determining what options for Linux servers should be made are based on the application needs. However, sometimes you already have the “server” and then are limited by what options exist. For example, if you have Scale-out or Enterprise POWER8 systems you can only virtualize with PowerVM. If you have L models you can virtualize both LE and BE distributions with PowerVM. DPLARs, PowerKVM, RHEV, or bare metal depends on which hypervisor you are running. If you have LC models you can run LE only distributions on bare metal or if you are virtualizing withPowerKVM then you can run BE or LE distributions.
If you are trying to determine which server best matches your application need you should start with the application..BE or LE?..then do you need optimum performance….then do you wish a virtualized environment. That should lead you to the correct server selection.
One final thought is that Little Endian distribution can only run on POWER8 systems.
- Summary table for POWER8 Systems:
LC models L models Scale-out models Enterprise models
BE or LE X (KVM, RHEV) X (KVM, RHEV, X (PowerVM) X (PowerVM)
LC models L models Scale-out models Enterprise models
BE or LE X (KVM, RHEV) X (KVM, RHEV, X (PowerVM) X (PowerVM)
(virtualized) PowerVM)
OPAL X (KVM, RHEV) X (KVM, RHEV) na na
pHyp na X (PowerVM) X (PowerVM) X (PowerVM)
PowerKVM – X (OPAL) X (OPAL) na na
OPAL X (KVM, RHEV) X (KVM, RHEV) na na
pHyp na X (PowerVM) X (PowerVM) X (PowerVM)
PowerKVM – X (OPAL) X (OPAL) na na
BE or LE
PowerVM - na X (pHyp) X X
PowerVM - na X (pHyp) X X
BE or LE
Bare Metal – X (OPAL) X (OPAL) na na
Bare Metal – X (OPAL) X (OPAL) na na
LE only
[{"Business Unit":{"code":"BU054","label":"Systems w\/TPS"},"Product":{"code":"HW1W1","label":"Power ->PowerLinux"},"Component":"","Platform":[{"code":"PF025","label":"Platform Independent"}],"Version":"","Edition":"","Line of Business":{"code":"","label":""}}]
UID
ibm16170277