While preparing for the GCP Professional Cloud Architect (PCA) exam, many candidates get confused by how to handle stateful workloads requiring shared storage at scale. In the real world, this is fundamentally a decision about managed shared file systems versus block storage, balancing performance, cost, and operational complexity. Let’s drill into a simulated scenario.
The Scenario #
NexLevel Gaming is a fast-growing global gaming startup that runs multiplayer game servers on Google Cloud. Their game servers are horizontally scalable but stateful: each server instance needs to read from and write to a shared filesystem with POSIX semantics for player state and session data. At peak load, their system must support sustained writes up to 100 MB/s to the shared storage. The infrastructure team wants to implement this on Google Cloud to ensure smooth game play and quick scaling.
Requirements #
Architect a solution that supports horizontal scaling of stateful instances with a shared POSIX filesystem, high write throughput (up to 100 MB/s), and operational simplicity.
The Options #
- A) Use a Persistent Disk attached individually to each instance.
- B) Use a Regional Persistent Disk attached individually to each instance.
- C) Create a Cloud Filestore instance and mount it to each game server instance.
- D) Create a Cloud Storage bucket and mount it on each instance using gcsfuse.
Correct Answer #
C) Create a Cloud Filestore instance and mount it to each game server instance.
The Architect’s Analysis #
Correct Answer #
Option C.
Step-by-Step Winning Logic #
Cloud Filestore is Google’s fully managed Network File System (NFS) that provides a POSIX-compliant shared filesystem accessible by multiple instances simultaneously. This matches the scenario’s requirement for consistent shared read/write access with POSIX semantics. It handles scale and throughput natively, supporting sustained high writes around 100 MB/s, making it much better suited than local or regional Persistent Disks that cannot be concurrently mounted in read-write mode by multiple instances.
This approach embraces SRE principles by reducing operational complexity—no custom replication or coordination logic needed. It also aligns with FinOps by balancing cost against operational risk and performance.
The Traps (Distractor Analysis) #
- Why not A? Attach a Persistent Disk per instance provides high-performance storage, but Persistent Disk cannot be attached read/write to multiple instances simultaneously. Each disk is single-attach in R/W mode, so this breaks the shared state requirement.
- Why not B? Regional Persistent Disks provide redundancy but still have the same limitation of single-attach for writes, so they don’t solve the shared POSIX storage need.
- Why not D? Mounting Cloud Storage buckets via gcsfuse gives shared access to objects but does not provide POSIX compliance, has high latencies, and performance is inadequate for 100 MB/s sustained writes. It also adds complexity through eventual consistency semantics and higher client-side caching.
The Architect Blueprint #
This diagram shows NexLevel’s horizontally scaled game server fleet accessing a shared Cloud Filestore instance.
Diagram Note: Each horizontally scaled game server instance mounts the shared Cloud Filestore NFS while serving live player sessions to maintain consistent game state.
The Decision Matrix #
| Option | Est. Complexity | Est. Monthly Cost | Pros | Cons |
|---|---|---|---|---|
| A) Persistent Disk per instance | Low | Low (PD storage + IOPS) | Simple, high IOPS per instance | Cannot share disk R/W, no POSIX shared FS |
| B) Regional Persistent Disk per instance | Low-Medium | Medium (regional PD costs) | Redundancy & durability | Still single-attach, no multi-instance write sharing |
| C) Cloud Filestore (NFS) | Medium | Medium-High (based on provisioned capacity) | Fully managed POSIX shared FS; scalable writes; simpler ops | Higher cost than PD; must provision capacity upfront |
| D) Cloud Storage + gcsfuse | Medium-High | Low (pay per use) | Cheap, highly durable object storage | No POSIX compliance; eventual consistency; poor write performance |
Real-World Practitioner Insight #
Exam Rule #
“For the exam, always pick Cloud Filestore when you see POSIX shared filesystem with concurrent write access and high throughput.”
Real World #
“In production, where cost is a strong concern and write throughput is low, teams might architect around replicated block storage or application-level coordination. However, for horizontally scaled stateful workloads with heavy consistent writes, Cloud Filestore is the recommended path for operational simplicity and reliability.”