Huawei Virtualization Platform Management and Usage · Topic 1 of 4
FusionCompute Compute Virtualization
Learn FusionCompute compute virtualization: clusters, live migration, HA, DRS, DPM, rule groups, memory overcommitment and CPU and memory QoS settings.
- 5 min
- The cluster is where HA, DRS, DPM and memory overcommitment are configured.
- HA restarts VMs after a host failure and needs shared storage; it is not zero downtime.
- DRS balances load with live migration; DPM saves power by emptying and powering off hosts.
What it is
Compute virtualization in FusionCompute turns the CPU and memory of many hosts into a shared pool. VMs draw vCPUs and memory from that pool, and the platform moves, restarts and balances them across hosts.
The unit that pools hosts is the cluster. Most compute features, such as HA, DRS and DPM, are switched on and tuned at the cluster level.
How it works
Clusters, hosts and VMs
A host is a physical server running CNA. A cluster groups hosts that share storage and networks so VMs can move freely between them.
Each VM gets vCPUs and virtual memory. The hypervisor schedules vCPUs onto physical CPU cores, so the total vCPUs in a cluster can exceed the physical cores.
VM live migration
Live migration moves a running VM from one host to another without stopping its services. Memory is copied to the destination while the VM runs, then a short final copy and switch completes the move.
| Migration type | What moves | Typical use |
|---|---|---|
| Live migration between hosts | The running VM's compute state | Host maintenance, load balancing |
| Storage migration | The VM's virtual disks to another datastore | Storage maintenance or rebalancing |
| Migration of compute and storage together | Both at once | Moving between hosts that do not share storage |
Live migration between hosts relies on both hosts seeing the same shared datastore. Hosts with different CPU generations may need the cluster's IMC (incompatible migration cluster) setting, which presents a common CPU feature set to VMs.
HA
HA (high availability) restarts VMs on other hosts in the cluster when their host fails. It is a restart, so the VM's services stop briefly and the guest OS boots again.
HA needs shared storage, because the new host must reach the VM's disks. Enough spare capacity must also exist on the surviving hosts, which is why HA settings can reserve resources for failover.
DRS and DPM
| Feature | Full name | What it does |
|---|---|---|
| DRS | Dynamic Resource Scheduling | Watches host load and live migrates VMs from busy hosts to quieter ones, by threshold |
| DPM | Distributed Power Management | When load is low, moves VMs off some hosts and powers those hosts off; powers them on again when load rises |
DPM works together with DRS, because it uses migration to empty hosts before power off. Both can run automatically or only give suggestions for the administrator to apply.
Rule groups
Rule groups tell DRS where VMs may or may not run.
| Rule | Effect |
|---|---|
| Keep VMs together | The VMs run on the same host, for example an application and its cache |
| Keep VMs mutually exclusive | The VMs run on different hosts, for example two nodes of a cluster, so one host fault does not stop both |
| VMs to hosts | A VM group runs, or does not run, on a chosen host group |
Memory overcommitment
Memory overcommitment lets the memory given to VMs exceed the physical memory of the host. It works because VMs rarely use all their memory at the same time. It is enabled for the cluster, and FusionCompute uses three techniques.
| Technique | How it frees memory |
|---|---|
| Memory sharing | VMs with identical memory pages, such as zero pages, share one physical copy until one writes to it |
| Memory ballooning (memory bubble) | A driver inside the VM, part of the Tools, claims idle guest memory and the host gives it to other VMs |
| Memory swapping | The host moves memory pages to disk storage; slowest, used when the others are not enough |
CPU and memory QoS
QoS settings control how VMs compete for resources.
| Setting | Meaning |
|---|---|
| Shares | Relative weight; a VM with more shares gets more when resources are contended |
| Reservation | A guaranteed minimum that the VM always receives |
| Limit | The maximum the VM can use even when resources are free |
Shares only matter under contention. When the host is idle, each VM can use what it needs up to its limit.
Running VMs can also get more vCPUs or memory without a restart through CPU and memory hot add, when the guest OS supports it.
A worked example
A cluster has four hosts and shared IP SAN storage. HA and DRS are enabled, and the two database VMs are in a "Keep VMs mutually exclusive" rule.
During the day, DRS sees host 1 above its load threshold and live migrates two web VMs to host 3. At night, DPM migrates VMs off host 4 and powers it off. When host 2 later suffers a power fault, HA restarts its VMs on the remaining hosts, and the rule keeps the two database VMs on different hosts.
How the exam asks about this
- "Which feature restarts VMs on another host when their host fails?" -> HA.
- "Which feature powers off lightly loaded hosts to save energy?" -> DPM, working with DRS.
- "Which memory overcommitment technique uses a driver inside the VM to reclaim idle memory?" -> Memory ballooning.
What to remember for the exam
- The cluster is where HA, DRS, DPM and memory overcommitment are configured.
- HA restarts VMs after a host failure and needs shared storage; it is not zero downtime.
- DRS balances load with live migration; DPM saves power by emptying and powering off hosts.
- Memory overcommitment uses memory sharing, memory ballooning and memory swapping.
- QoS uses shares, reservation and limit; reservation is a guaranteed minimum, limit is a maximum.
Live migration and HA are different. Live migration moves a running VM with no service stop; HA restarts a VM after its host has already failed.
Common mistakes
Calling HA seamless. The guest OS boots again on the new host, so services are briefly down and memory state is lost.
Treating shares as a guarantee. Shares are relative weights used under contention. Only a reservation guarantees a minimum amount.
Enabling DPM without DRS. DPM depends on migration to empty a host before it powers off, so it works together with DRS.
Check yourself
Which three techniques does FusionCompute use for memory overcommitment?
Memory sharing, memory ballooning and memory swapping, with swapping the slowest and used last.
What does a "Keep VMs mutually exclusive" rule do?
It makes DRS keep the selected VMs on different hosts, so one host fault cannot stop all of them.
Why does HA need shared storage?
The healthy host that restarts the VM must be able to reach the VM's virtual disks, which are on the shared datastore.
Topic 1 of 4 in Huawei Virtualization Platform Management and Usage
Test yourself on Huawei Virtualization Platform Management and Usage
The full Huawei H13-511 question bank has 600 questions, with timed practice exams and an explanation for every answer.
This study guide is independent and is not affiliated with, endorsed by, or sponsored by Huawei. Vendor names, certification names and exam codes are trademarks of their respective owners and are used for identification only.