-
Feature
-
Resolution: Unresolved
-
Critical
-
None
-
None
-
Product / Portfolio Work
-
False
-
-
False
-
Not Selected
-
-
100% To Do, 0% In Progress, 0% Done
This is a follow up to VIRTSTRAT-81
Feature Overview
OpenShift Virtualization is now enabling higher workload density by default and is in turn making the use of your hardware more efficient.
Goals
- Allow customers to leverage their existing hardware. Today, due to the memory overcommit differences between existing hypervisors i.e. (1.5:1) and OCPv (1:1 by default) has lead to the situation, where the existing hardware did not have enough capacity anymore in order to run all the VMs it was able to run before - while it could, because the workloads had a low utilization.
- Today enabling memory overcommit via HCO, and provisioning SWAP are to actions on the customer. After this feature, this work will be done automatically, and transparently for all deployments.
- Simplify testing by making this feature (higher density) part of the default flow.
- Customers
Requirements
Phase | Requirement | Notes | isMvp? |
---|---|---|---|
1 | Enable the descheduler with the correct profiles (incl mem) by default | It's a MUST in order to leverage memory across the cluster efficiently | Yes |
1 | Review, update, create a dashboard to monitor memory over committment | Yes | |
1 | Add memory support to descheduler | Today the utilization profile does not take memory into account | Yes |
2 | Allow SWAP on LUN | CNV-64266 | No |
2 | Explore: Move swap provision to WASP | to simplify the picture | Yes |
2 | Explore: Provision a small number of swap by default | We need to define the size of swap and the location, LUN is requested, makes sense IMHO | Yes |
Last | Enable memory overcommit by default | We need to define on the factor | Yes |
Future | Explore when to use compressed swap | No |
(Optional) Use Cases
How will the user interact with this feature? It will be almost transparent to users.
Out of Scope
Fix swap in OCP.
This is still a long term goal, but out of scope of this feature.
Eventually this feature would consume swap from OCP once it lands.
Background, and strategic fit
Two main reasons:
1. Improve testabilty by making it the default
2. Improve competitiveness by enabling memory overcommit by default
Assumptions
- <your text here>
Risks
CNV will have to own things that otherwise belong to the platform (SWAP and descheduler configuration).
Customer Considerations
We had 2 customers asking for making higher density simpler. (At this point in time)
Documentation Considerations
We want to create a dedicated section to describe higher density from a holistic pov.
User Experience Considerations
No UXD support needed
- depends on
-
RFE-8045 Enabled autoSizingReserved by default
-
- Backlog
-
-
VIRTSTRAT-537 Memory aware load balancing
-
- In Progress
-
- is blocked by
-
VIRTSTRAT-541 Memory compression for memory over committment
-
- New
-
- is related to
-
CNV-64266 Allow SWAP on SAN provided LUNs
-
- New
-
- relates to
-
CNV-17404 Allow scheduler to rearrange the cluster to allow certain actions to succeed
-
- In Progress
-
-
COS-2231 Support zram & swap on RHCOS
-
- Closed
-
-
VIRTSTRAT-139 Allow for larger VM to start in a resource limited cluster through workload rebalancing
-
- New
-
-
VIRTSTRAT-79 Increase workload density by adding memory oversubscription
-
- In Progress
-
- split from
-
VIRTSTRAT-80 Node level optimization to Increase workload density by adding memory oversubscription
-
- Closed
-
-
VIRTSTRAT-81 Add Agent to create swap on nodes until Node Swap Enabalement is done
-
- Closed
-