Cloud Virtualization
Cloud Virtualization
This domain tests the compute virtualization layer that underpins cloud: hypervisors, virtual switching, device models, and performance tuning for VNFs. You should know KVM/QEMU, libvirt, virtio, SR-IOV, DPDK, NUMA, and live migration constraints.
Core rules
- Type 1 hypervisors run on bare metal; KVM is a Linux kernel module, while ESXi and Xen are Type 1 examples. Type 2 runs on a host OS. QEMU emulates devices, KVM accelerates CPU virtualization, and libvirt manages domains.
- Virtio is the paravirtualized I/O framework, including virtio-net and virtio-blk. vhost-user moves the data plane to a userspace vSwitch; DPDK poll-mode drivers bypass the kernel; SR-IOV passes a VF directly to the VM for near-native throughput.
- NUMA placement, CPU pinning, and hugepages reduce jitter. Overcommitting vCPU or memory can cause noisy-neighbor effects and violates deterministic NFV SLAs.
- Live migration requires shared storage or block migration, compatible CPU features, and network reachability. SR-IOV VFs and some passthrough devices complicate or block migration.
Worked example
A vSRX must forward 10 Gbps with low jitter. Choose SR-IOV VFs for data interfaces, pin vCPUs to cores on the same NUMA node, and reserve 1 GB hugepages. Reasoning: SR-IOV avoids vSwitch overhead; pinning and hugepages reduce scheduling and TLB misses. Virtio with a kernel vSwitch is simpler but may not meet the throughput target.
Traps
- Assuming SR-IOV VMs can live-migrate freely. VF identity and host placement matter.
- Overcommitting memory on NFV hosts. Ballooning and swapping hurt forwarding.
- Mixing NUMA nodes for vCPU and memory, causing cross-socket latency.