4.4 Cloud Storage Classes & Persistent Disk Types
Key Takeaways
- Cloud Storage classes trade storage cost against retrieval cost and minimum duration: Standard (none), Nearline (30 days, ~$0.01/GB retrieval), Coldline (90 days, ~$0.02/GB), Archive (365 days, ~$0.05/GB).
- Deleting or moving an object before its class's minimum storage duration still bills for the full minimum duration — an early-deletion gotcha the exam tests directly.
- Object Lifecycle Management automates class transitions and deletions based on object age or version count, avoiding manual cleanup scripts.
- Persistent Disk types range from pd-standard (HDD, cheapest) through pd-balanced (console default) and pd-ssd to pd-extreme (independently provisioned IOPS); Hyperdisk is required on newest machine series.
- Regional Persistent Disk synchronously replicates across two zones for zone-outage resilience; Local SSD is high-performance but does not durably survive most VM stop/restart events.
Why This Section Matters
The exam guide closes out "Planning and configuring a cloud solution" with choosing storage options, explicitly naming Persistent Disk types and Cloud Storage's Standard, Nearline, Coldline, and Archive classes. This is a numbers-heavy topic — minimum storage durations, retrieval fees, and IOPS figures — and the ACE exam loves to test the exact thresholds, because getting them wrong in production means an expensive surprise on a customer's bill.
Cloud Storage Classes (Object Storage)
Cloud Storage is Google Cloud's object storage service, and every bucket/object is assigned a storage class that trades access cost against storage cost based on how often the data is expected to be read.
| Storage class | Minimum storage duration | Retrieval cost | Typical access pattern |
|---|---|---|---|
| Standard | None | None | Frequently accessed ("hot") data, active websites, streaming |
| Nearline | 30 days | ~$0.01/GB | Accessed roughly once a month or less; backups |
| Coldline | 90 days | ~$0.02/GB | Accessed roughly once a quarter or less; disaster recovery |
| Archive | 365 days | ~$0.05/GB | Accessed roughly once a year or less; long-term compliance retention |
Storage price per GB decreases as you move from Standard toward Archive, while retrieval cost increases in the opposite direction — the trade-off is always cheaper to store, more expensive to retrieve. A critical exam detail: if you delete, overwrite, or move an object before its class's minimum storage duration has elapsed, you are still billed as if it had been stored for the full minimum duration. A Coldline object deleted after 10 days is billed for the full 90-day minimum, not the 10 days it actually sat in the bucket.
Object Lifecycle Management automates this trade-off: you define rules (based on object age, number of newer versions, or other conditions) that automatically transition an object to a cheaper class or delete it outright, so a team doesn't have to run manual cleanup scripts. A classic lifecycle rule: move objects from Standard to Nearline after 30 days, to Coldline after 90 days, and to Archive after 365 days, deleting anything older than 7 years for compliance.
Persistent Disk & Hyperdisk Types (Block Storage)
Persistent Disk (PD) is Compute Engine's durable, network-attached block storage — the virtual "hard drives" attached to VMs.
| Disk type | Media | Performance tier | Best for |
|---|---|---|---|
| pd-standard | HDD-backed | Lowest cost, best for sequential I/O | Batch processing, large sequential reads/writes, cost-sensitive workloads |
| pd-balanced | SSD-backed | Balance of cost and performance (console default) | General-purpose boot disks and application workloads |
| pd-ssd | SSD-backed | Higher, consistent IOPS/throughput than balanced | Databases and I/O-intensive workloads |
| pd-extreme | SSD-backed | Highest PD tier; lets you separately provision target IOPS | Enterprise databases, SAP HANA |
| Hyperdisk (Balanced/Extreme/Throughput) | Next-generation block storage | Independently provisioned capacity, IOPS, and throughput; required on newest machine series (e.g., C4, N4) | Highest-performance workloads and machine types where classic Persistent Disk is not offered |
Two additional facts the exam expects you to know:
- Zonal vs. regional Persistent Disk: a zonal PD lives in and is replicated only within a single zone (cheaper, lower latency, but lost if that zone fails); a regional Persistent Disk is synchronously replicated across two zones in a region, protecting the disk's data against a single-zone outage at a higher cost — pick regional PD when a scenario requires resilience to a zone failure without relying purely on application-level replication.
- Local SSD is physically attached to the host server (not network-attached like PD), offering very high IOPS and low latency, but its data does not survive a VM stop, restart onto different hardware, or host maintenance event without live migration — it is appropriate only for caches, scratch space, or temporary processing data, never as a system of record.
Realistic Exam Scenario
A compliance team must retain audit logs for exactly seven years, accessed maybe once during that entire period if legal discovery is required, with the absolute lowest possible storage cost. The correct choice is the Archive storage class combined with an Object Lifecycle Management rule that deletes objects after 7 years — Archive's 365-day minimum and highest retrieval fee are irrelevant here because access is essentially never expected, and the storage-cost savings dominate.
Now contrast a database scenario: an enterprise SAP HANA deployment needs guaranteed, independently tunable IOPS beyond what a standard SSD Persistent Disk offers. The answer is pd-extreme (or Hyperdisk Extreme on newer machine series) — only these tiers let you provision IOPS as a separate, explicit parameter rather than deriving it from disk size.
Common Traps
- Choosing Nearline for data accessed "a few times a year" — that access pattern is Coldline or Archive territory; Nearline is for roughly monthly access, and picking it for rarer access wastes money on a higher per-GB storage rate than necessary.
- Forgetting the early-deletion penalty — a scenario describing an object moved to Coldline and deleted after only a few weeks still incurs the full 90-day minimum charge, a fact the exam likes to test as a "gotcha."
- Assuming Local SSD is a durable choice for database files — Local SSD data does not persist across most VM stop/restart or maintenance events, so any scenario requiring durability rules it out in favor of a Persistent Disk type.
A healthcare provider stores patient imaging files that are almost never accessed again after 90 days but must be retained for regulatory reasons for 10 years, with cost as the primary concern. Which storage class best fits the bulk of this data after the initial 90-day window?
A team provisions a Compute Engine VM that will only cache intermediate build artifacts during a CI pipeline run and can tolerate total data loss if the VM restarts, in exchange for the lowest possible latency. Which disk type is most appropriate?