COA Performance Optimization 3 — Questions and Answers
Question 1: Which Nova configuration option allows CPU overcommit ratios to be set, enabling more vCPUs than physical cores on a host?
- cpu_allocation_ratio (Correct answer)
- vcpu_pin_set
- cpu_dedicated_set
- reserved_host_cpus
Correct answer: cpu_allocation_ratio
cpu_allocation_ratio multiplies reported physical CPUs by the given factor, allowing Nova to schedule more vCPUs than physically available.
Question 2: What Neutron mechanism driver provides the best network I/O performance for bare-metal-like workloads using kernel bypass?
- SR-IOV with ml2 sriovnicswitch (Correct answer)
- Open vSwitch with VXLAN
- Linux bridge with VLAN
- OVN with Geneve
Correct answer: SR-IOV with ml2 sriovnicswitch
SR-IOV passes virtual functions directly to instances, bypassing the hypervisor network stack for near-native throughput and lowest latency.
Question 3: In a Ceph-backed Glance deployment, which image format eliminates the conversion overhead during Nova instance boot?
- raw (Correct answer)
- qcow2
- vmdk
- vhd
Correct answer: raw
Raw images are directly usable by Ceph RBD without any conversion, so Nova boots instances from them without a CPU-intensive format conversion step.
Question 4: Which Heat resource type should be used to implement auto-scaling policies based on Ceilometer/Aodh alarms?
- OS::Heat::ScalingPolicy
- OS::Nova::ServerGroup
- OS::Heat::AutoScalingGroup (Correct answer)
- OS::Aodh::Alarm
Correct answer: OS::Heat::AutoScalingGroup
OS::Heat::AutoScalingGroup manages a group of resources that can grow or shrink in response to OS::Heat::ScalingPolicy actions triggered by alarms.
Question 5: What is the primary performance benefit of enabling NUMA topology awareness in Nova when scheduling instances?
- Instances are placed so their vCPUs and memory reside on the same NUMA node, minimizing cross-node memory latency (Correct answer)
- Instances get dedicated physical CPUs exclusively
- SR-IOV NICs are automatically attached
- Hugepages are automatically allocated
Correct answer: Instances are placed so their vCPUs and memory reside on the same NUMA node, minimizing cross-node memory latency
NUMA-aware scheduling ensures that instance vCPUs access memory on the same NUMA node, avoiding expensive cross-node memory transfers.
Question 6: Which oslo.cache backend provides the highest performance for Keystone token caching in a multi-node deployment?
- dogpile.cache.memcached (Correct answer)
- dogpile.cache.memory
- dogpile.cache.dbm
- dogpile.cache.null
Correct answer: dogpile.cache.memcached
Memcached provides a shared, high-performance distributed cache accessible by all Keystone nodes, unlike the in-process memory backend which is per-node only.
Question 7: An operator notices that Nova API response times degrade as the instance count grows. Which database optimization best addresses this?
- Adding indexes on frequently queried columns like project_id, host, and status in the nova_api database (Correct answer)
- Increasing nova-api worker count only
- Switching to SQLite for lower overhead
- Enabling nova-conductor pooling
Correct answer: Adding indexes on frequently queried columns like project_id, host, and status in the nova_api database
Database query performance degrades with large tables without proper indexes; adding indexes on project_id, host, and status speeds up common Nova API queries.
COA Performance Optimization 3
Which Nova configuration option allows CPU overcommit ratios to be set, enabling more vCPUs than physical cores on a host?