Skip to main content
  1. Home
  2. >
  3. GCP
  4. >
  5. PCA
  6. >
  7. High-Throughput Web Apps—Compute vs Storage | GCP PCA

High-Throughput Web Apps—Compute vs Storage | GCP PCA

Jeff Taakey
Author
Jeff Taakey
21+ Year Enterprise Architect | Multi-Cloud Architect & Strategist.

While preparing for the GCP Professional Cloud Architect (PCA) exam, many candidates wrestle with choosing the right compute and database services for high-throughput workloads that demand real-time querying at scale. In the real world, this is fundamentally a decision about balancing serverless managed platforms versus VM-based autoscaling, and fast key-value storage versus analytical data warehousing. Let’s drill into a simulated scenario.

The Scenario
#

FinLedger is a rapidly scaling global fintech startup that collects financial transaction data from millions of users worldwide. Their core web service ingests data at a blistering rate—up to 500,000 requests per second during peak trading hours. The platform must store this data in a way that enables real-time querying on exact attribute matches to immediately detect fraud patterns. However, due to market downtime, there are lengthy periods when no data is ingested. As a startup, FinLedger is keen to optimize operational overhead and keep cloud costs as low as possible without sacrificing reliability or scalability.

Key Requirements
#

Design the cloud service architecture that can efficiently receive large spikes of transactional data, store it for real-time exact-match queries, operate under intermittent workload conditions, and minimize total cost of ownership.

The Options
#

  • A) Cloud Run and BigQuery
  • B) Cloud Run and Cloud Bigtable
  • C) Compute Engine autoscaling managed instance group and BigQuery
  • D) Compute Engine autoscaling managed instance group and Cloud Bigtable

Correct Answer
#

B) Cloud Run and Cloud Bigtable


The Architect’s Analysis
#

Correct Answer
#

Option B: Cloud Run and Cloud Bigtable

Step-by-Step Winning Logic
#

  • Managed serverless ingress with Cloud Run allows autoscaling from zero to massive concurrency with no preprovisioning or VM management, reducing operational toil (a key SRE principle).
  • Cloud Bigtable supports ultra-low latency, exact-match, high-throughput queries using a NoSQL wide-column model ideal for time series/transactional data. It scales elastically and suits the access pattern.
  • Cloud Run’s pay-per-use model fits the intermittent traffic pattern perfectly, avoiding cost waste during idle periods.
  • BigQuery is designed for analytical queries, not low-latency exact-match lookups; pairing it with Cloud Run leads to latency and cost inefficiencies.
  • Compute Engine instance groups require managing VM lifecycle and potentially preprovisioning capacity, increasing ops complexity and cost risk.

The Traps (Distractor Analysis)
#

  • Option A (Cloud Run + BigQuery): BigQuery’s analytic engine does not support millisecond-level exact-match lookups efficiently.
  • Option C (Compute Engine + BigQuery): More operational overhead managing VMs; BigQuery still not ideal for real-time querying.
  • Option D (Compute Engine + Bigtable): While Bigtable is suitable, managing Compute Engine autoscaling groups adds complexity and cost risks, especially with traffic volatility.

The Architect Blueprint
#

Mermaid Diagram illustrating managed ingestion and fast storage flow.

graph TD User([User Client]) --> |Requests| CloudRun["Cloud Run (autoscaling)"] CloudRun --> |Write Data| Bigtable[Cloud Bigtable] QueryService --> |Exact Match Queries| Bigtable style CloudRun fill:#4285F4,stroke:#333,color:#fff style Bigtable fill:#0F9D58,stroke:#333,color:#fff

Diagram Note: User requests hit Cloud Run which elastically scales to handle bursts, writing transactional data into Cloud Bigtable for fast exact-match retrieval by downstream services.

The Decision Matrix
#

Option Est. Complexity Est. Monthly Cost Pros Cons
A Low Medium to High Serverless compute; fully managed BigQuery too slow and costly for lookups
B Low Low Managed serverless + Bigtable scaling Requires understanding Bigtable schema
C Medium Medium to High Flexible VM control VM ops toil; BigQuery not for real-time
D Medium Medium Bigtable low latency VM management overhead; complexity spikes

Real-World Practitioner Insight
#

Exam Rule
#

“For the PCA exam, favor managed serverless platforms like Cloud Run when facing unpredictable or spiky workloads combined with real-time data ingestion needs.”

Real World
#

“FinOps best practice dictates pairing pay-per-use serverless compute with an elastic NoSQL store like Bigtable for massive scale and low operational burden, especially when workloads fluctuate dramatically.”

GCP Professional Cloud Architect Drills

Design, develop, and manage robust, secure, and scalable Google Cloud solutions.