VM Placement & Scheduling¶

VM Pinning¶

VM pinning prevents a VM from being automatically migrated by maintenance mode evacuation or Dynamic Placement. Pinned VMs remain on their assigned host unless manually moved by an administrator.

Use Cases¶

• VMs with host-level hardware dependencies (GPU passthrough, USB devices)

• License-bound VMs that must run on specific hosts

• VMs with local storage that cannot be migrated

• System VMs (cluster management agents)

Behavior of Pinned VMs¶

Placement Strategy¶

Each VM has a placement strategy that determines how HPE Morpheus Software handles its host assignment:

Dynamic Placement¶

Dynamic Placement automatically balances VM placement across cluster hosts based on resource utilization. It runs during each cluster sync cycle.

Failover Behavior (HA)¶

When a host becomes unreachable, HPE Morpheus Software detects the failure during the cluster health check and takes the following actions:

1. Detection — The host is marked unreachable after failing connectivity checks during a cluster sync cycle.

2. VM identification — All VMs that were running on the failed host are identified.

3. Placement evaluation — For each affected VM with Auto placement strategy, HPE Morpheus Software selects a target host with sufficient available memory and CPU capacity.

4. Restart — VMs are restarted on their assigned target hosts. VMs with Pinned placement strategy are NOT automatically failed over and remain offline until the original host recovers or an administrator intervenes.

After the original host recovers:

• VMs with Failover placement strategy that were moved to a different host are automatically migrated back to their preferred (original) host once it is healthy and has capacity.

• VMs with Auto placement strategy remain on their current host — use Dynamic Placement or manual migration to rebalance if desired.

Affinity Group Enforcement During Failover¶

During failover, affinity rules are evaluated:

• Keep Together — If one VM in a “Keep Together” group is failed over to a new host, other VMs in the group are also migrated to the same host (capacity permitting).

• Keep Separate — VMs in a “Keep Separate” group are placed on different surviving hosts to maintain anti-affinity.

Modes¶

Dynamic Placement respects:

• VM pinning (pinned VMs are never moved)

• Maintenance mode flags (hosts in maintenance are excluded)

• Host-VM Group affinity and anti-affinity rules

• Available memory thresholds

Configuring Dynamic Placement¶

Dynamic Placement settings are configured when editing a cluster:

1. Navigate to Infrastructure ‣ Clusters and select the cluster

2. Click Edit

3. Enable the Dynamic Placement toggle to activate DRS for the cluster

4. Set the Aggressiveness level to control how actively VMs are rebalanced

5. Click Save

When Dynamic Placement is disabled (the default), the cluster operates in Failover Only mode and no automatic rebalancing occurs.

Aggressiveness Levels¶

The aggressiveness level controls how readily HPE Morpheus Software migrates VMs to rebalance the cluster. Higher aggressiveness results in tighter balancing but more frequent live migrations. Three levels are available:

Threshold Parameters¶

Each aggressiveness level sets a group of threshold parameters that govern the DRS algorithm:

How the Algorithm Works¶

During each cluster sync cycle, Dynamic Placement evaluates the cluster as follows:

1. Cluster balance check — Calculate the coefficient of variation (CV) for memory and CPU across all hosts. If both values are below the CV Threshold, the cluster is considered balanced and no action is taken.

2. Identify overloaded hosts — Hosts whose memory usage exceeds the mean multiplied by the Imbalance Threshold are flagged.

3. Select VMs to move — On each overloaded host, eligible VMs are evaluated for migration. A VM is eligible if it uses the Auto placement strategy, has no local storage or assigned devices, is powered on, is not in an active backup, is not in an affinity group, and has not been migrated within the cooldown period.

4. Select target hosts — Target hosts must have memory and CPU usage below the mean multiplied by the Target Low Threshold. The move is also validated to ensure it would not push the target host above the Imbalance Threshold.

5. Validate improvement — The algorithm simulates the move and confirms it would reduce the cluster’s standard deviation by at least the Min Std Dev Improvement percentage.

6. Execute migrations — Approved moves are executed up to the Max Moves Per Cycle limit.

Host-VM Groups (Affinity/Anti-Affinity)¶

Host-VM Groups define rules that control which hosts a VM may run on. These are configured at the cluster level and enforced during provisioning, migration, and Dynamic Placement operations.

For detailed instructions on creating and managing Host-VM Groups, see Host-VM Groups.

Dynamic Placement and Affinity Enforcement¶

When Dynamic Placement runs with dynamic placement enabled, it evaluates affinity groups:

• VMs in an affinity group that are not on their preferred host are migrated back (if the preferred host is online and has capacity)

• VMs in an anti-affinity group that violate the rule (running on the same host as another group member) are migrated to separate hosts

Memory Overcommit¶

By default, HPE Morpheus Software treats VM memory allocations as hard reservations — a VM can only be placed on a host with enough free physical memory to fully back the allocation. Memory overcommit allows you to allocate more memory to VMs than the host physically has available, under the assumption that not all VMs will use their full allocation simultaneously.

Configuring Memory Overcommit¶

Memory overcommit is set per host from the host detail page:

1. Navigate to the host detail page (Infrastructure ‣ Clusters > select cluster > Hosts tab > select host)

2. Click Edit

3. Set the Overcommit Percent field to the desired value

4. Click Save

How Overcommit Affects Placement¶

When HPE Morpheus Software selects a host for a new VM, it calculates available memory as:

available = (host physical memory × overcommit percent) - reserved memory - allocated VM memory

A host is only eligible for placement if this calculated available memory exceeds the new VM’s memory requirement. This means overcommit expands the pool of eligible hosts but does not bypass the placement check entirely.

Dynamic Placement also uses the overcommit-adjusted memory when evaluating rebalancing decisions.

Overcommit Alarms¶

When the total allocated VM memory on a host exceeds the host’s physical memory capacity (adjusted by the overcommit percent), HPE Morpheus Software raises a Host Memory Over-Committed alarm. This alarm indicates that the host is approaching or has exceeded its overcommit threshold and may experience performance degradation if VMs simultaneously consume their full allocations.