GCP Professional Data Engineer Practice Question

Your team operates a millisecond-sensitive trading platform on Google Compute Engine. Each request must read and update a 1 KB position record about 60 000 times per second, and the 95th-percentile end-to-end latency must stay below 2 ms. Records must survive a single-zone outage. The current design writes the state to a regional SSD persistent disk attached to every VM, but the application is now hitting disk IOPS limits and latency SLOs. Which redesign best meets the performance and durability targets while minimizing ongoing cost?

Store the position records in a Standard-Tier Cloud Memorystore (Redis) instance that replicates across zones and takes periodic RDB snapshots to Cloud Storage.
Replace each regional SSD persistent disk with four Local SSDs striped at the guest-OS level and export an hourly backup to Cloud Storage for durability.
Increase each regional SSD persistent disk to 6 TB to lift the disk's IOPS quota and distribute traffic across more Compute Engine instances.
Migrate the position records to a three-node Cloud Bigtable cluster accessed through the HBase client library running on every VM.

Report Issue

Answer Description

Cloud Memorystore for Redis keeps all data in memory and delivers sub-millisecond latency, comfortably meeting the 2 ms SLO. The Standard Tier provides a primary-replica deployment in separate zones, so the dataset remains available if one zone fails. Because Memorystore pricing is based on provisioned memory rather than per-node reservation (Bigtable) or provisioned storage size (persistent disks), it is typically less expensive than scaling SSD persistent disks or adding Bigtable nodes for this workload.

Local SSDs would meet latency, but the data is erased if a VM is stopped or crashes; periodic backups cannot guarantee zero data loss after a zone failure. Bigtable scales to very high throughput but generally delivers single-digit millisecond latency-insufficient for the 2 ms requirement-and node-based pricing makes it costlier for a 60 000-QPS workload. Simply enlarging regional SSD persistent disks raises IOPS limits but still involves network storage with several-millisecond round trips, so the latency problem remains and storage costs rise proportionally with capacity.

Ask Bash

Bash is our AI bot, trained to help you pass your exam. AI Generated Content may display inaccurate information, always double-check anything important.