Live vCPU & RAM Scaling

Overview

Live vCPU/RAM scaling allows users to adjust compute resources on running instances without downtime. As an administrator, you configure which flavors support live scaling by setting hot-add extra specs that define the resource envelope.

Xloud-Developed — Live vCPU/RAM Scaling is developed by Xloud and ships with XAVS / XPCI.

Prerequisites

Admin credentials sourced from admin-openrc.sh
Understanding of flavor extra specs and NUMA topology

Key Capabilities

Bidirectional CPU Scaling

Add and remove vCPUs on a running instance. Xloud is the only enterprise cloud platform that supports vCPU hot-remove through the management UI — other platforms support add-only.

Three Memory Methods

Balloon — fast cooperative resize within existing ceiling. DIMM Hotplug — 256 MB-aligned persistent modules. virtio-mem — 4 MB granularity with bidirectional add/remove support.

CPU Topology Awareness

Automatic thread alignment for SMT (hyperthreading) hosts. Five-method detection cascade ensures correct topology mapping across all CPU generations.

Configure a Live-Scalable Flavor

Before users can live-scale instances, you must create flavors with hot-add extra specs that define the min/max resource envelope.

Dashboard
CLI

Open Flavor Management

Navigate to Admin > Compute > Flavors.

Create or edit a flavor

Click Create Flavor (or select an existing flavor and click Update Metadata).Set the base resources — these are the starting values when an instance is launched:

Field	Example	Description
Name	`m1.large-scalable`	Descriptive name indicating live scaling is enabled
vCPUs	`4`	Starting vCPU count
RAM (MB)	`8192`	Starting RAM in MB
Root Disk (GB)	`80`	Root disk (unchanged by live scaling)

Add hot-add extra specs

In the flavor’s Extra Specs panel, add the following keys:

Key	Example Value	Description
`hw:cpu_max_sockets`	`16`	Maximum CPU sockets (sets vCPU ceiling)
`hw:cpu_max_cores`	`1`	Cores per socket (keep at 1 unless using NUMA)
`hw:cpu_max_threads`	`1`	Threads per core
`hw:mem_max_mb`	`65536`	Maximum RAM in MB (64 GB ceiling)
`hw:mem_page_size`	`any`	Set to `any` or `large` for memory balloon resizing

The ceiling for vCPUs = hw:cpu_max_sockets x hw:cpu_max_cores x hw:cpu_max_threads. For a flavor starting at 4 vCPUs scaling to 16, set hw:cpu_max_sockets=16, hw:cpu_max_cores=1, hw:cpu_max_threads=1.

Source your credentials file to authenticate with the Xloud platform:

Load credentials

source admin-openrc.sh

Download the OpenRC file from Xloud Dashboard → Project → API Access → Download OpenStack RC File.

Create a live-scalable flavor (4-16 vCPU, 8-64 GB RAM)

openstack flavor create \
  --vcpus 4 \
  --ram 8192 \
  --disk 80 \
  --public \
  m1.large-scalable

Set hot-add extra specs — vCPU envelope

openstack flavor set m1.large-scalable \
  --property hw:cpu_max_sockets=16 \
  --property hw:cpu_max_cores=1 \
  --property hw:cpu_max_threads=1

Set hot-add extra specs — RAM envelope

openstack flavor set m1.large-scalable \
  --property hw:mem_max_mb=65536 \
  --property hw:mem_page_size=any

Verify extra specs are set

openstack flavor show m1.large-scalable -c properties

The flavor shows hw:cpu_max_sockets, hw:mem_max_mb in its properties — it is now live-scaling enabled.

Resource Bounds Reference

The allowed range for live scaling is determined by the flavor’s base values and hot-add extra specs:

Resource	Minimum	Maximum	Configured By
vCPUs	Flavor base vCPU count	`hw:cpu_max_sockets x hw:cpu_max_cores x hw:cpu_max_threads`	Flavor extra specs
RAM	Flavor base RAM (MB)	`hw:mem_max_mb`	Flavor extra spec

The minimum is the base flavor value — users cannot scale below the flavor’s original allocation via live scaling. To permanently reduce resources below the base, use a standard flavor resize (requires reboot).

API Reference

Live scaling is available via the Xloud Compute API extension:

Adjust vCPU and RAM on a running instance

curl -X POST \
  "https://api.<your-domain>/compute/v2.1/os-xloud-adjust/<server-id>" \
  -H "X-Auth-Token: $OS_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "xloud_adjust": {
      "current_vcpus": 8,
      "current_memory_mb": 16384
    }
  }'

Field	Type	Description
`current_vcpus`	integer	New desired vCPU count (within flavor envelope)
`current_memory_mb`	integer	New desired RAM in MB (within flavor envelope)

You can pass one or both fields — omit the field you do not want to change.

Get current live resource state

curl -X GET \
  "https://api.<your-domain>/compute/v2.1/servers/<server-id>/xloud-status" \
  -H "X-Auth-Token: $OS_TOKEN"

NUMA Considerations

When an instance has a NUMA topology defined (via hw:numa_nodes), vCPU changes must keep each NUMA node balanced:

A 2-NUMA-node instance with 4 vCPUs has 2 vCPUs per node
Scaling to 6 vCPUs is valid (3 per node); scaling to 5 vCPUs is not (uneven split)
Scaling to 8 vCPUs is valid (4 per node)

Check NUMA topology of a flavor

openstack flavor show m1.large-scalable -c properties | grep numa

For instances without an explicit hw:numa_nodes extra spec, the hypervisor uses a single NUMA cell and any valid vCPU count within the envelope is accepted.

Troubleshooting

Live vCPU/RAM button not visible for users

Cause: The instance’s flavor does not have hot-add extra specs configured.Fix: Add hw:cpu_max_sockets and/or hw:mem_max_mb to the flavor. Existing instances on that flavor will immediately become live-scalable — no reboot required.

Check if flavor has hot-add specs

openstack flavor show <flavor-name> -c properties

Scaling request rejected — value out of range

Cause: The requested vCPU or RAM value exceeds the flavor’s configured maximum.Fix: Either adjust the requested value, or update the flavor’s max specs:

Increase vCPU ceiling to 32

openstack flavor set <flavor-name> --property hw:cpu_max_sockets=32

vCPU change rejected due to NUMA topology

Cause: The requested vCPU count cannot be evenly distributed across the instance’s NUMA nodes.Fix: Choose a vCPU count that divides evenly across NUMA nodes. For a 2-NUMA instance: use 2, 4, 6, 8, 10…

Guest OS does not show new vCPUs after scaling up

Cause: Older Linux kernels (before 4.15) require manual CPU onlining.Fix: Run inside the guest:

Bring all hotplugged CPUs online

for cpu in /sys/devices/system/cpu/cpu*/online; do echo 1 > $cpu; done

Memory scaling appears to have no effect

Cause: The guest OS memory balloon driver (virtio_balloon) may not be loaded.Fix: Verify and load:

Check and load balloon driver

lsmod | grep virtio_balloon
modprobe virtio_balloon

For persistence, add virtio_balloon to /etc/modules.

CPU pinned instances cannot be live-scaled

Cause: Instances with hw:cpu_policy=dedicated (CPU pinning) do not support live vCPU scaling — this is by design to prevent pinning conflicts.Fix: Use hw:cpu_policy=shared (default) for instances that need live scaling. CPU-pinned instances must use standard flavor resize instead.

Next Steps

User Guide: Live Scaling

End-user guide for scaling vCPU and RAM from the Dashboard

Flavor Management

Create and manage flavors with extra specs

Advanced Features

CPU pinning, NUMA, huge pages, vTPM, and GPU passthrough

Compute Admin Guide

Return to the Compute Administration Guide index

Core Services

Other Services

Live vCPU & RAM Scaling

Overview

Key Capabilities

Bidirectional CPU Scaling

Three Memory Methods

CPU Topology Awareness

Configure a Live-Scalable Flavor

Open Flavor Management

Create or edit a flavor

Add hot-add extra specs

Resource Bounds Reference

API Reference

NUMA Considerations

Troubleshooting

Next Steps

User Guide: Live Scaling

Flavor Management

Advanced Features

Compute Admin Guide

Core Services

Other Services

​Overview

​Key Capabilities

Bidirectional CPU Scaling

Three Memory Methods

CPU Topology Awareness

​Configure a Live-Scalable Flavor

Open Flavor Management

Create or edit a flavor

Add hot-add extra specs

​Resource Bounds Reference

​API Reference

​NUMA Considerations

​Troubleshooting

​Next Steps

User Guide: Live Scaling

Flavor Management

Advanced Features

Compute Admin Guide

Overview

Key Capabilities

Configure a Live-Scalable Flavor

Resource Bounds Reference

API Reference

NUMA Considerations

Troubleshooting

Next Steps