Follow up for Intel SST-PP spec
This patch fixes the comments on https://review.opendev.org/#/c/651459/9 Change-Id: If06a019cf97aca48949054b331b526f4d9bf6cc1
This commit is contained in:
Madhuri Kumari
parent
b40310a2a7
commit
973a1833b7
|
|
@ -10,19 +10,18 @@ Add Intel® Speed Select Support
|
|||
|
||||
https://storyboard.openstack.org/#!/story/2005390
|
||||
|
||||
Multiple type of servers are needed to handle diverse workloads. Purchasing
|
||||
Multiple types of servers are needed to handle diverse workloads. Purchasing
|
||||
and managing these servers introduces complexity and increases total cost of
|
||||
ownership(TCO).
|
||||
Intel Speed Select Technology(SST)[1] is a collection of features that improves
|
||||
performance and optimize TCO by providing more control over CPU performance.
|
||||
With Intel SST, one server can do more which means same server can be used to
|
||||
run different workloads. One of the feature is Intel Speed Select
|
||||
Technology-Performance Profile (SST-PP) allows configuring the CPU to run at
|
||||
3 distinct operating points or profiles.
|
||||
performance and optimizes TCO by providing more control over CPU performance.
|
||||
With Intel SST, one server can do more which means the same server can be used
|
||||
to run different workloads. One of the feature is Intel Speed Select
|
||||
Technology-Performance Profile (SST-PP) which allows configuring the CPU to run
|
||||
at 3 distinct operating points or profiles.
|
||||
|
||||
With Intel SST-PP, one can set the desired core count and their base and turbo
|
||||
frequency. It allows to set the number of cores and their frequencies which can
|
||||
help to tune the server to specific workload performance.
|
||||
frequencies which can help to tune the server to specific workload performance.
|
||||
|
||||
Problem description
|
||||
===================
|
||||
|
|
@ -56,8 +55,8 @@ Proposed change
|
|||
===============
|
||||
|
||||
Intel SST-PP can be set over IPMI. Each configuration level has its
|
||||
own hexa raw code that server understands. Ironic sends this code to the server
|
||||
via IPMI to set the desired SST-PP level.
|
||||
own hexa raw code that the server understands. Ironic sends this code to the
|
||||
server via IPMI to set the desired SST-PP level.
|
||||
|
||||
We will map these configurations to traits that Ironic understands.
|
||||
|
||||
|
|
@ -72,21 +71,21 @@ Scheduling
|
|||
|
||||
The solution to support Intel SST-PP is first enable scheduling the request
|
||||
for baremetal instance deployment on the nodes that supports it. We set the
|
||||
desired configuration as the trait in our flavor.
|
||||
desired configuration as the trait in our flavor::
|
||||
|
||||
$ openstack flavor set --property \
|
||||
trait:CUSTOM_INTEL_SPEED_SELECT_CONFIG_2=required baremetal
|
||||
|
||||
Now, we also need to update the Ironic node's trait with the supported
|
||||
configuration levels.
|
||||
configuration levels::
|
||||
|
||||
$ openstack --os-baremetal-api-version 1.37 baremetal node add trait node-0 \
|
||||
$ openstack baremetal node add trait node-0 \
|
||||
CUSTOM_INTEL_SPEED_SELECT_CONFIG_BASE CUSTOM_INTEL_SPEED_SELECT_CONFIG_1 \
|
||||
CUSTOM_INTEL_SPEED_SELECT_CONFIG_2
|
||||
|
||||
|
||||
Now, when user sends a request to boot a node with the ``baremetal`` flavor.
|
||||
Placement API service will select the Ironic node that supports Intel SST-PP.
|
||||
Now, when user sends a request to boot a node with the ``baremetal`` flavor,
|
||||
placement API service will select the Ironic node that supports Intel SST-PP.
|
||||
|
||||
Provisioning
|
||||
------------
|
||||
|
|
@ -95,8 +94,8 @@ The Intel SST-PP needs to be set via IPMI before powering on the node
|
|||
in the process of provisioning. Ironic API service receives the desired
|
||||
configuration in ``node.instance_info.traits`` in the boot request. Ironic will
|
||||
then run the deploy template's step matching the trait. The deploy template
|
||||
will specify the new ``configure_intel_speedselect`` step which configure
|
||||
the Intel SST-PP configuration and then power on the node.
|
||||
will specify the new ``configure_intel_speedselect`` step which configures
|
||||
the Intel SST-PP configuration level and then powers on the node.
|
||||
|
||||
Ironic will need below details to configure Intel SST-PP on the node:
|
||||
|
||||
|
|
@ -105,8 +104,8 @@ Ironic will need below details to configure Intel SST-PP on the node:
|
|||
is set by admins in the flavor's trait they want to boot the baremetal node
|
||||
with. Nova in turn updates the Ironic's node information with the trait.
|
||||
|
||||
* No. of socket: This is the no. of sockets per CPU. Setting Intel SST-PP needs
|
||||
to be done for every socket.
|
||||
* Number of sockets: This is the number of sockets per CPU. Setting Intel
|
||||
SST-PP needs to be done for every socket.
|
||||
|
||||
Both these information can be provided as an argument to the
|
||||
``configure_intel_speedselect`` deploy step.
|
||||
|
|
@ -156,40 +155,40 @@ Driver API impact
|
|||
-----------------
|
||||
|
||||
* Add a new hardware type ``IntelIPMIHardware`` to support Intel SST-PP
|
||||
enabled hardwares.
|
||||
enabled servers.
|
||||
|
||||
.. code-block:: python
|
||||
.. code-block:: python
|
||||
|
||||
class IntelIPMIHardware(IPMIHardware):
|
||||
class IntelIPMIHardware(IPMIHardware):
|
||||
|
||||
* Add a new management interface ``IntelIPMIManagement`` to manage configuring
|
||||
Intel SST-PP on the servers. This class will have a new deploy step
|
||||
``configure_intel_speedselect`` to configure Intel SST-PP on the nodes. This
|
||||
will also be enabled as a clean step so that it can be used to reset
|
||||
configuration while node cleaning.
|
||||
* Add a new management interface ``IntelIPMIManagement`` to manage the
|
||||
configuration of Intel SST-PP on the servers. This class will have a new
|
||||
deploy step ``configure_intel_speedselect`` to configure Intel SST-PP on
|
||||
the nodes. This will also be enabled as a clean step so that it can be used
|
||||
to reset the configuration while node cleaning.
|
||||
|
||||
|
||||
.. code-block:: python
|
||||
.. code-block:: python
|
||||
|
||||
class IntelIPMIManagement(ipmitool.IPMIManagement):
|
||||
@base.deploy_step(priority=200, argsinfo={
|
||||
'intel_speedselect_config': {
|
||||
'description': (
|
||||
"Intel SST-PP configuration."
|
||||
),
|
||||
'required': True
|
||||
},
|
||||
'socket_count': {
|
||||
'description': (
|
||||
"No. of sockets."
|
||||
)
|
||||
}
|
||||
})
|
||||
def configure_intel_speedselect(self, task, **kwargs):
|
||||
return None
|
||||
class IntelIPMIManagement(ipmitool.IPMIManagement):
|
||||
@base.deploy_step(priority=200, argsinfo={
|
||||
'intel_speedselect_config': {
|
||||
'description': (
|
||||
"Intel SST-PP configuration."
|
||||
),
|
||||
'required': True
|
||||
},
|
||||
'socket_count': {
|
||||
'description': (
|
||||
"No. of sockets."
|
||||
)
|
||||
}
|
||||
})
|
||||
def configure_intel_speedselect(self, task, **kwargs):
|
||||
return None
|
||||
|
||||
Following is table for ``intel_speedselect_config`` value
|
||||
corresponding each deploy template::
|
||||
The following table contains the ``intel_speedselect_config`` values
|
||||
for the proposed deploy templates::
|
||||
|
||||
-----------------------------------------------------------------------
|
||||
| Deploy Template | intel_speedselect_config |
|
||||
|
|
@ -218,8 +217,8 @@ None
|
|||
Other end user impact
|
||||
---------------------
|
||||
|
||||
Users will have to update the traits with required Intel SST-PP configuration
|
||||
level in flavor.
|
||||
Users will have to update the traits with the desired Intel SST-PP
|
||||
configuration level in the corresponding flavors.
|
||||
|
||||
Scalability impact
|
||||
------------------
|
||||
|
|
|
|||
Loading…
Reference in New Issue