[openstack-ansible] : GPU computes on openstack-ansible
Hi Team Openstack-Ansible Openstack Victoria Release Ubuntu 20.04 I need to add computes having GPU on them. So I need recommendations on the procedure to add them and also what all configurations through openstack-ansible are required for nova_api, nova_scheduler on control node nova_api containers and compute node nova_compute configurations? Also does anyone has experience using and installing openstack service : cyborg for GPU management. Regards Rambo
Hey, We actually had Nvidia GPUs being provisioned and supplied to end users, but we did that using vGPU feature. So passing through GPUs with PCI -passthrough will be slightly different. And the cyborg is really another way of doing things. So having more specifics on design details would be helpful, as there are quite some ways of passing GPUs to end users. On Fri, 22 Nov 2024, 18:11 Rambo Rambo, <ram.ramb2412@gmail.com> wrote:
Hi Team
Openstack-Ansible Openstack Victoria Release Ubuntu 20.04
I need to add computes having GPU on them. So I need recommendations on the procedure to add them and also what all configurations through openstack-ansible are required for nova_api, nova_scheduler on control node nova_api containers and compute node nova_compute configurations?
Also does anyone has experience using and installing openstack service : cyborg for GPU management.
Regards Rambo
Hi Dmitriy We are considering both options at the moment GPU passthrough and Cyborg and at moment we don't intend to use split sample physical GPU into multiple vGPUs. I could see we need to configure below in the nova configuration on the control nodes and computes for the passthrough option: nova-api [pci] passthrough_whitelist: alias: [] enabled_filters: nova-compute [pci] alias = And Create flavor with below property and attach this flavor with the instances. --property "pci_passthrough:alias" For cyborg: Can I get the sample configuration that can help me to start exploring this option, Regards Rambo On Fri, Nov 22, 2024 at 8:56 PM Dmitriy Rabotyagov <noonedeadpunk@gmail.com> wrote:
Hey,
We actually had Nvidia GPUs being provisioned and supplied to end users, but we did that using vGPU feature. So passing through GPUs with PCI -passthrough will be slightly different. And the cyborg is really another way of doing things. So having more specifics on design details would be helpful, as there are quite some ways of passing GPUs to end users.
On Fri, 22 Nov 2024, 18:11 Rambo Rambo, <ram.ramb2412@gmail.com> wrote:
Hi Team
Openstack-Ansible Openstack Victoria Release Ubuntu 20.04
I need to add computes having GPU on them. So I need recommendations on the procedure to add them and also what all configurations through openstack-ansible are required for nova_api, nova_scheduler on control node nova_api containers and compute node nova_compute configurations?
Also does anyone has experience using and installing openstack service : cyborg for GPU management.
Regards Rambo
Hi Team Please confirm if it is possible to deploy cyborg service using the openstack-ansible. I am checking /etc/openstack_deploy/conf.d/ and didn't find any example file for cyborg. if its possible can I get some reference for the same. Thanks in advance. Regards Rambo On Fri, Nov 22, 2024 at 9:17 PM Rambo Rambo <ram.ramb2412@gmail.com> wrote:
Hi Dmitriy
We are considering both options at the moment GPU passthrough and Cyborg and at moment we don't intend to use split sample physical GPU into multiple vGPUs.
I could see we need to configure below in the nova configuration on the control nodes and computes for the passthrough option:
nova-api
[pci]
passthrough_whitelist:
alias:
[]
enabled_filters:
nova-compute
[pci]
alias =
And Create flavor with below property and attach this flavor with the instances.
--property "pci_passthrough:alias"
For cyborg: Can I get the sample configuration that can help me to start exploring this option,
Regards Rambo
On Fri, Nov 22, 2024 at 8:56 PM Dmitriy Rabotyagov < noonedeadpunk@gmail.com> wrote:
Hey,
We actually had Nvidia GPUs being provisioned and supplied to end users, but we did that using vGPU feature. So passing through GPUs with PCI -passthrough will be slightly different. And the cyborg is really another way of doing things. So having more specifics on design details would be helpful, as there are quite some ways of passing GPUs to end users.
On Fri, 22 Nov 2024, 18:11 Rambo Rambo, <ram.ramb2412@gmail.com> wrote:
Hi Team
Openstack-Ansible Openstack Victoria Release Ubuntu 20.04
I need to add computes having GPU on them. So I need recommendations on the procedure to add them and also what all configurations through openstack-ansible are required for nova_api, nova_scheduler on control node nova_api containers and compute node nova_compute configurations?
Also does anyone has experience using and installing openstack service : cyborg for GPU management.
Regards Rambo
Hey, Unfortunately, we do not have support for Cyborg as of today due to historically limited interest in the project. Though it might be interesting to add the support for the project in case you come up with a role code for its installation and configuration. We can assist you with details on how such a process usually looks like and provide possible help on the way. With that said, you will still be able to use the role with OSA until it's being reviewed and merged, which is the case with any custom component you want to add to your deployment. вт, 26 нояб. 2024 г. в 10:53, Rambo Rambo <ram.ramb2412@gmail.com>:
Hi Team
Please confirm if it is possible to deploy cyborg service using the openstack-ansible. I am checking /etc/openstack_deploy/conf.d/ and didn't find any example file for cyborg. if its possible can I get some reference for the same. Thanks in advance.
Regards Rambo
On Fri, Nov 22, 2024 at 9:17 PM Rambo Rambo <ram.ramb2412@gmail.com> wrote:
Hi Dmitriy
We are considering both options at the moment GPU passthrough and Cyborg and at moment we don't intend to use split sample physical GPU into multiple vGPUs.
I could see we need to configure below in the nova configuration on the control nodes and computes for the passthrough option:
nova-api
[pci]
passthrough_whitelist:
alias:
[]
enabled_filters:
nova-compute
[pci]
alias =
And Create flavor with below property and attach this flavor with the instances.
--property "pci_passthrough:alias"
For cyborg: Can I get the sample configuration that can help me to start exploring this option,
Regards Rambo
On Fri, Nov 22, 2024 at 8:56 PM Dmitriy Rabotyagov <noonedeadpunk@gmail.com> wrote:
Hey,
We actually had Nvidia GPUs being provisioned and supplied to end users, but we did that using vGPU feature. So passing through GPUs with PCI -passthrough will be slightly different. And the cyborg is really another way of doing things. So having more specifics on design details would be helpful, as there are quite some ways of passing GPUs to end users.
On Fri, 22 Nov 2024, 18:11 Rambo Rambo, <ram.ramb2412@gmail.com> wrote:
Hi Team
Openstack-Ansible Openstack Victoria Release Ubuntu 20.04
I need to add computes having GPU on them. So I need recommendations on the procedure to add them and also what all configurations through openstack-ansible are required for nova_api, nova_scheduler on control node nova_api containers and compute node nova_compute configurations?
Also does anyone has experience using and installing openstack service : cyborg for GPU management.
Regards Rambo
Hi Rambo, Dmitriy We haven't added the support for Cyborg in openstack-ansible project, but kolla-ansible is available. Our team are investing to add the support for Cyborg in openstack-ansible. Regards Alex -----邮件原件----- 发件人: Dmitriy Rabotyagov <noonedeadpunk@gmail.com> 发送时间: 2024年11月26日 18:03 收件人: Rambo Rambo <ram.ramb2412@gmail.com> 抄送: openstack-discuss@lists.openstack.org 主题: Re: [openstack-ansible] : GPU computes on openstack-ansible Hey, Unfortunately, we do not have support for Cyborg as of today due to historically limited interest in the project. Though it might be interesting to add the support for the project in case you come up with a role code for its installation and configuration. We can assist you with details on how such a process usually looks like and provide possible help on the way. With that said, you will still be able to use the role with OSA until it's being reviewed and merged, which is the case with any custom component you want to add to your deployment. вт, 26 нояб. 2024 г. в 10:53, Rambo Rambo <ram.ramb2412@gmail.com>:
Hi Team
Please confirm if it is possible to deploy cyborg service using the openstack-ansible. I am checking /etc/openstack_deploy/conf.d/ and didn't find any example file for cyborg. if its possible can I get some reference for the same. Thanks in advance.
Regards Rambo
On Fri, Nov 22, 2024 at 9:17 PM Rambo Rambo <ram.ramb2412@gmail.com> wrote:
Hi Dmitriy
We are considering both options at the moment GPU passthrough and Cyborg and at moment we don't intend to use split sample physical GPU into multiple vGPUs.
I could see we need to configure below in the nova configuration on the control nodes and computes for the passthrough option:
nova-api
[pci]
passthrough_whitelist:
alias:
[]
enabled_filters:
nova-compute
[pci]
alias =
And Create flavor with below property and attach this flavor with the instances.
--property "pci_passthrough:alias"
For cyborg: Can I get the sample configuration that can help me to start exploring this option,
Regards Rambo
On Fri, Nov 22, 2024 at 8:56 PM Dmitriy Rabotyagov <noonedeadpunk@gmail.com> wrote:
Hey,
We actually had Nvidia GPUs being provisioned and supplied to end users, but we did that using vGPU feature. So passing through GPUs with PCI -passthrough will be slightly different. And the cyborg is really another way of doing things. So having more specifics on design details would be helpful, as there are quite some ways of passing GPUs to end users.
On Fri, 22 Nov 2024, 18:11 Rambo Rambo, <ram.ramb2412@gmail.com> wrote:
Hi Team
Openstack-Ansible Openstack Victoria Release Ubuntu 20.04
I need to add computes having GPU on them. So I need recommendations on the procedure to add them and also what all configurations through openstack-ansible are required for nova_api, nova_scheduler on control node nova_api containers and compute node nova_compute configurations?
Also does anyone has experience using and installing openstack service : cyborg for GPU management.
Regards Rambo
participants (3)
-
Alex Song (宋文平)
-
Dmitriy Rabotyagov
-
Rambo Rambo