Loading...

Change Owns to Parent Ofs

Set start and due date...

Xporter

XML

Word

Printable

Details

Type: Feature
Priority: Should have
Fix Version/s: PI22
Component/s: None
Labels:

ARTs:

SRCnet
Benefit hypothesis:

Hide

Ideally, all SRCNet services can run on Kubernetes across all SRCNet sites.

It is believed that many SRCNet sites are already running OpenStack.

This feature is looking at evaluating if Azimuth can help OpenStack sites run performant Kubernetes clusters, within a minimal local effort required.

In addition, more consistency across SRCNet sites might help simplify rolling out new versions of SRCNet services.

Show
Ideally, all SRCNet services can run on Kubernetes across all SRCNet sites. It is believed that many SRCNet sites are already running OpenStack. This feature is looking at evaluating if Azimuth can help OpenStack sites run performant Kubernetes clusters, within a minimal local effort required. In addition, more consistency across SRCNet sites might help simplify rolling out new versions of SRCNet services.
Acceptance criteria:

Hide

AC1: Document what HPC and Cloud CoP members do today around support (or not) Kubernetes, including details about how that runs on OpenStack or something else.

AC2: Demo providing Kubernetes using Azimuth at a site other than Cambridge Arcus, currently RAL STFC cloud are offering to do that.

AC3: (stretch) Get one or more non-UK sites testing if the Azimuth installation docs are sufficient to get Azimuth running at their site, ideally updating the docs with any gaps we find.

AC4: (stretch) Ideally at least one SRCNet site is able to test and demo Azimuth's existing built-in support for at least one of: NVidia GPUs and RDMA networking using Mellanox ConnectEx5 (or above) NICs.

Show
AC1: Document what HPC and Cloud CoP members do today around support (or not) Kubernetes, including details about how that runs on OpenStack or something else. AC2: Demo providing Kubernetes using Azimuth at a site other than Cambridge Arcus, currently RAL STFC cloud are offering to do that. AC3: (stretch) Get one or more non-UK sites testing if the Azimuth installation docs are sufficient to get Azimuth running at their site, ideally updating the docs with any gaps we find. AC4: (stretch) Ideally at least one SRCNet site is able to test and demo Azimuth's existing built-in support for at least one of: NVidia GPUs and RDMA networking using Mellanox ConnectEx5 (or above) NICs.
Feature Points:
2
Initial Size:
2
WSJF:
0
Agile Teams:

Team_CORAL, Team_TEAL
Due Sprint:
Sprint 4
Story Point Burn-up:
Overdue:
Overdue

Requirement Status:

PI23 - UNCOVERED
Labels_MIRO:
SRCNet0.1 Teal-D operations-and-infrastructure site-provisioning

Description

Some SRCNet services are expected to be best run on Kubernetes. For SRCNet sites that are already looking to use OpenStack, it would be good if we can share efforts on good practices for creating a performant, maintainable and upgradable Kubernetes cluster running on OpenStack. There are currently many ways to achieve that aim, all with different trade off depending on the use case.

Azimuth can provide performant on demand and upgradable Kubernetes clusters, using regular project level user access to an OpenStack project. (There is a work in progress effort for a Magnum driver to expose this stack via Magnum APIs, but that does require cloud changes, that can not be assumed).

Details on how to install Azimuth on an OpenStack are already available. Additional feedback on how repeatable this process is would be very welcome:
https://stackhpc.github.io/azimuth-config/

We are proposing that we work with the UK STFC Cloud team to try and get this up and running and following all required security policies. They do already make some use of the capi-helm-charts directly. It would be good to see if there are other SRCNet sites where sharing effort around Kubernetes on OpenStack would be helpful. It would be good to compare the approach with other efforts globally, and start to build consensus on how best to share efforts in this area.

In particular it would be good to cover areas such as:

Baremetal cluster support (via Ironic+Nova)
GPU support (VMs with passthrough, not vGPU)
RDMA support, Ethernet and Infini-band

Attachments

Issue Links

mentioned in: Page Loading...; Page Loading...; Page Loading...; Page Loading...

Structure

Activity

People

Assignee:: Bolton, Rosie

Reporter:: Watson, Duncan

Votes:: 0 Vote for this issue

Watchers:: 1 Start watching this issue

Feature Progress

Story Point Burn-up: (43.14%)

Feature Estimate: 2.0

	Issues	Story Points
To Do	6	20.0
In Progress	5	9.0
Complete	4	22.0
Total	15	51.0

Dates

Created:: 16/Feb/24 8:31 PM

Updated:: 6 days ago 12:09 PM

Due Sprint Date:: 07/May/24

Collecting, evaluating and sharing K8s on OpenStack good practice