[Kimchi-devel] RFC – Hypervisor Performance Manager for s390x (KVM IBM z Systems)

Thu Dec 1 15:15:51 UTC 2016

Hi Chandra,

On 11/23/2016 08:47 AM, Chandra Shekhar Reddy Potula wrote:
>
> Proposal of introducing a new plugin – Hypervisor Performance 
> Manager(HPM) for s390x (KVM IBM z Systems).
>
>

I'd say to we understand better on the requirements to then identify if 
a new plugin is really needed or it can be contained on Kimchi.

1) Are there similar tools for x86 and ppc?

> *Abstract*
>
> z Systems Hypervisor Performance Manager (zHPM) can be used to bring a 
> goal-oriented approach to the performance management of a hypervisor. 
> zHPM dynamically manages resource allocation when contention occurs 
> based on user-specified policies associated with performance objectives.
>
> *Introduction about HPM:*
>
> zHPM provides the following capabilities:
>
> ·APIs to enable a virtualization management solution for fully 
> operating zHPM.
>
> ·These APIs are provided using RESTful web services. The zHPM APIs 
> provide the following functions:
>

> –Definition of workload resource groups and performance policy.
>

Could you elaborate more on it?

How do the things work? What is it needed from user and how that 
information is used?

> –Collection of virtual server and hypervisor performance metrics 
> related to understanding of performance goal achievement.
>

What is a virtual server? Is it a hypervisor host?
Is the idea behind it to collect statistics on how the virtual machines 
are affecting the host in a global view? Like condensing the individual 
stats we have today for each virtual machine in one place?

> –Operational controls.
>

Could you elaborate more?

> ·A command line interface that provides similar functions as the zHPM 
> APIs.
>

Why a new command line if the user can use zHPM directly? Or they can't 
do that ?

> ·Collection of hypervisor level and virtual server level performance 
> metrics to help set up performance goals, and support virtual server 
> CPU management.
>

Some new concepts here. =)

What is hypervisor leval and virtual server level?

> ·Management of CPU resource allocation to virtual servers in order to 
> meet goals
>

Is virtual server a virtual machine? We already have CPU allocation for 
each virtual machine. Is the idea to condensing that into one view?

>
> Workload resource groups
>
> A workload resource group is a collection of virtual servers that host 
> a business application. The workload resource group provides a 
> "management view" of these virtual servers. By defining a workload 
> resource group you can manage hypervisor resources based on the 
> requirements of the business applications deployed on the KVM for IBM 
> z Systems hypervisor.
>
> Performance policies
>
> In zHPM, a performance policy defines goals for virtual servers in a 
> workload resource group. You can define performance policies and 
> express a business goal and an importance level for the virtual 
> servers that are part of a workload resource group. zHPM then monitors 
> and manages hypervisor resources allocated to the virtual servers 
> based on these performance policies.
>
> Service class
>
> the service class a virtual server is associated with.
>
> *Use Cases:*
>
> New plugin Hypervisor Performance Manager(HPM) UI toprovide the 
> following operations support:
>
>
> 1.Performance Policies
>
>         a.List
>
>         b.Add/Update/Remove
>
>         c.Import/Export
>
> 2.Workload Resource Group (WRG) configuration
>
>         a.List WRG’s
>
>         b.Get/Create/Edit/Delete WRG
>
>         c.Add/Remove Virtual Server from WRG
>
> 3.Service Classes
>
>         a.Details of Service Class
>
>         b.Fetch or update Virtual Server information
>
> 4.Virtual Server CPU Management
>
>         a.Get/Enable/Disable CPU management
>
>         b.Report dynamic CPU resource adjustments
>
> REST API’s Supported by HPM.
>
> 1.Workload Resource Group
>
> a.GET /zhpm/wsapi/v1/workload-resource-groups
>
> b.GET /zhpm/wsapi/v1/workload-resource-groups/detail
>
> c.GET /zhpm/wsapi/v1/workload-resource-groups/{wrg-resource-id}
>
> d.POST /zhpm/wsapi/v1/workload-resource-groups
>
> e.DELETE /zhpm/wsapi/v1/workload-resource-groups/{wrg-resource-id}
>
> f.POST /zhpm/wsapi/v1/workload-resource-groups/{wrg-resource-id}/actions
>
> 2.Performance Policies
>
> a.POST /zhpm/wsapi/v1/workload-resource-groups/{wrg-resource-id}
>
> b.POST /zhpm/wsapi/v1/workload-resource-groups/default
>
> c.POST /zhpm/wsapi/v1/workload-resource-groups/{wrg-resource-ids
>
> d.GET /zhpm/wsapi/v1/workload-resource-groups/{wrg-resource-id}
>
> e.GET /zhpm/wsapi/v1/workload-resource-groups/detail
>
> 3.Virtual Servers
>
> a.GET /zhpm/wsapi/v1/virtual-servers
>
> b.GET /zhpm/wsapi/v1/virtual-servers/{vs-resource-id}
>
> c.POST /zhpm/wsapi/v1/virtual-servers/{vs-resource-id}
>
> d.GET /zhpm/wsapi/v1/virtual-servers/detail
>
> 4.Metrics
>
> a.GET /zhpm/wsapi/v1/metrics/raw/hypervisor
>
> b.GET /zhpm/wsapi/v1/metrics/calculated/hypervisor
>
> c.GET /zhpm/wsapi/v1/metrics/raw/workload-resource-groups
>
> d.GET /zhpm/wsapi/v1/metrics/calculated/workload-resource-groups
>
> e.GET /zhpm/wsapi/v1/metrics/raw/workload-resource-groups/default
>
> f.GET /zhpm/wsapi/v1/metrics/calculated/workload-resource-groups/default
>
> g.GET 
> /zhpm/wsapi/v1/metrics/raw/workload-resource-groups/{wrg-resource-id}
>
> a.GET 
> /zhpm/wsapi/v1/metrics/calculated/workload-resource-groups/{wrg-resource-id}
>
> 5.CPU Management
>
> a.GET /zhpm/wsapi/v1/mgmt-levels
>
> b.POST /zhpm/wsapi/v1/mgmt-levels
>
> c.GET /zhpm/wsapi/v1/dynamic-resource-adjustments
>
> 6.Memory Management
>
> TBD
>
> The following code changes will be made in support of plugin:
>
> 1.Plugin infrastructure
>
> 2.Backend python code support for Kimchi/Ginger for the REST API 
> development
>
> 3.Frontend HTML5/JavaScript code.
>
>
> More details to be followed like detailed use cases, REST API and 
> mock-ups ...
>
>
> Further documentation can be found @ 
> http://www.ibm.com/support/knowledgecenter/SSNW54_1.1.1/com.ibm.kvm.v111.admin/part2.htm?cp=linuxonibm
>
>
>  Please feel free to comment !!!
>
>

Not sure I understand the whole idea but for me it seems to be an 
overall host view when running virtual machines.
So probably, it should be part of Ginger plugin and not a new plugin or 
part of Kimchi.

Please, correct me if I am wrong.

> Thanks and Regards
>
> Chandra
>
>
>
> _______________________________________________
> Kimchi-devel mailing list
> Kimchi-devel at ovirt.org
> http://lists.ovirt.org/mailman/listinfo/kimchi-devel

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ovirt.org/pipermail/kimchi-devel/attachments/20161201/1bac5df5/attachment.html>