Performance Optimization Flashcards
7 cards from real COA practice questions. Tap to flip, then mark Knew It or Still Learning — missed cards come back until you master them.
Read the first 7 Performance Optimization flashcards as text
Which Glance configuration option improves image download performance by enabling multiple parallel data streams from the backend?
Answer: show_image_direct_url = True
show_image_direct_url exposes the backend storage URL directly so Nova can fetch images in parallel from the backend rather than proxying through Glance.
Which OpenStack component should be used to pre-warm image caches on compute nodes before a large batch of instance launches?
Answer: nova image-cache (image pre-caching API)
Nova's image pre-caching API allows operators to push specific images to compute node local caches ahead of demand, avoiding slow first-boot image downloads.
What is the purpose of enabling hugepages (1GB or 2MB) for OpenStack instances running memory-intensive workloads?
Answer: Hugepages reduce TLB misses and page table overhead, improving memory throughput for large working-set applications
Hugepages allow the CPU TLB to cover more memory with fewer entries, dramatically reducing TLB miss penalties for workloads with large memory footprints.
Which Nova scheduler weigher causes instances to be spread across compute hosts by default to maximize available resources per host?
Answer: RAMWeigher with a negative multiplier
RAMWeigher with ram_weight_multiplier set to a positive value ranks hosts by most available RAM, effectively spreading instances across nodes.
In an OVN-based Neutron deployment, which feature replaces traditional L3 agents on each compute node to reduce north-south latency?
Answer: OVN distributed virtual routing (DVR)
OVN implements distributed virtual routing natively within the OVN control plane, routing traffic locally on the compute node without an external L3 agent hop.
Which metric reported by the Nova compute service best indicates CPU steal time that degrades instance performance on a busy hypervisor?
Answer: nova diagnostics cpu0_time
nova diagnostics returns per-CPU metrics including cpu0_time which, compared to wall-clock time, reveals how much CPU time the instance is losing to steal.
Which action best reduces control plane overhead when managing 10,000+ instances in a single OpenStack cell?
Answer: Splitting the deployment into multiple Nova cells (CellsV2) to partition the instance database and message queue load
CellsV2 partitions instances, compute nodes, and message queues into isolated cells, preventing any single database or RabbitMQ broker from becoming a bottleneck.