CTO Systems Architecture & Infrastructure 3 — Questions and Answers
Question 1: A CTO must choose a networking model for a hybrid cloud environment connecting on-premises data centers to two public clouds. Which approach provides the lowest latency and most consistent bandwidth?
- Site-to-site VPN over public internet
- Dedicated private interconnects (e.g., AWS Direct Connect, Azure ExpressRoute) to each cloud provider (Correct answer)
- SD-WAN overlaying consumer broadband links
- Public internet with QUIC-based transport optimization
Correct answer: Dedicated private interconnects (e.g., AWS Direct Connect, Azure ExpressRoute) to each cloud provider
Dedicated private interconnects bypass the public internet, offering guaranteed bandwidth SLAs and sub-10ms latency not achievable over VPN or broadband.
Question 2: Which design pattern best prevents cascading failures when a downstream service experiences high latency in a synchronous microservices call chain?
- Retry with exponential backoff only
- Timeout combined with a circuit breaker that opens after a failure threshold (Correct answer)
- Increasing thread pool sizes on all callers
- Switching to a synchronous gRPC protocol
Correct answer: Timeout combined with a circuit breaker that opens after a failure threshold
A circuit breaker stops calls to a degraded service after repeated failures, shedding load and preventing upstream thread exhaustion while the service recovers.
Question 3: An architect proposes a CQRS pattern for a high-read e-commerce catalog. What infrastructure component is most critical to make the read side performant without overloading the write store?
- A synchronous two-phase commit between read and write stores
- An asynchronous event stream (e.g., Kafka) projecting write events to a denormalized read model (Correct answer)
- A read replica of the write database with synchronous replication
- A global distributed cache with no TTL
Correct answer: An asynchronous event stream (e.g., Kafka) projecting write events to a denormalized read model
CQRS typically uses an event stream to asynchronously update an optimized read model, decoupling read and write scaling concerns.
Question 4: A CTO evaluates moving from a colocation facility to a hyperscaler. Which Total Cost of Ownership factor is most frequently underestimated in early cloud migration business cases?
- Virtual machine hourly rates
- Egress bandwidth charges for data leaving the cloud provider's network (Correct answer)
- Storage capacity costs
- Compute reserved instance pricing
Correct answer: Egress bandwidth charges for data leaving the cloud provider's network
Cloud providers charge significant fees for outbound data transfer, which compounds quickly for high-traffic applications and is rarely modeled accurately upfront.
Question 5: When designing a disaster recovery strategy, what metric defines the maximum amount of data an organization can afford to lose, expressed in time?
- RTO — Recovery Time Objective
- RPO — Recovery Point Objective (Correct answer)
- MTTR — Mean Time to Recovery
- SLA — Service Level Agreement
Correct answer: RPO — Recovery Point Objective
RPO (Recovery Point Objective) specifies the maximum tolerable data loss window, directly determining backup frequency and replication lag limits.
Question 6: Which storage architecture is most appropriate for a machine learning training pipeline that reads large sequential datasets from thousands of small files concurrently?
- NFS v3 mounted on each training node
- Object storage (e.g., S3) with parallel multi-part reads and file aggregation (Correct answer)
- Local NVMe SSD on a single training host
- iSCSI block storage with synchronous writes
Correct answer: Object storage (e.g., S3) with parallel multi-part reads and file aggregation
Object storage scales throughput horizontally with parallel reads and eliminates small-file penalties when datasets are pre-aggregated (e.g., TFRecord, Parquet).
Question 7: A CTO is assessing the risk of a vendor providing a proprietary container orchestration platform. Which technical mitigation most preserves future portability?
- Negotiate a longer SLA with the vendor
- Build application manifests against the Kubernetes API and avoid vendor-specific CRDs in the critical path (Correct answer)
- Use vendor-specific deployment tooling but document all steps
- Run workloads on bare metal to avoid vendor platforms entirely
Correct answer: Build application manifests against the Kubernetes API and avoid vendor-specific CRDs in the critical path
Kubernetes-standard manifests can be redeployed on any conformant cluster; reliance on proprietary CRDs creates lock-in at the API surface level.
CTO Systems Architecture & Infrastructure 3
A CTO must choose a networking model for a hybrid cloud environment connecting on-premises data centers to two public clouds.
Which approach provides the lowest latency and most consistent bandwidth?