Storage Tier Overview
Google Cloud Storage offers four storage classes, each designed for different access frequency and durability requirements. Standard storage is optimised for frequently accessed data with no minimum storage duration. Nearline targets data accessed less than once per month, with a 30-day minimum. Coldline is designed for quarterly access with a 90-day minimum. Archive serves long-term retention with a 365-day minimum and the lowest storage cost but highest retrieval fees.
All four tiers provide identical durability (99.999999999% or eleven nines) and are backed by the same underlying infrastructure. The differences lie in storage pricing, retrieval costs, and access latency characteristics. For this benchmark, we created test buckets in the us-central1 region with uniform bucket-level access enabled.
Upload Throughput
We measured upload throughput using the gsutil tool with parallel composite uploads enabled, transferring datasets ranging from 1 GB to 100 GB. Tests ran from an n2-standard-8 instance in the same region as the storage bucket to eliminate network variability. Each test was repeated five times; the table shows median throughput.
| Dataset Size | Standard (MB/s) | Nearline (MB/s) | Coldline (MB/s) | Archive (MB/s) |
|---|---|---|---|---|
| 1 GB (100 files) | 485 | 482 | 478 | 476 |
| 10 GB (1,000 files) | 1,240 | 1,235 | 1,228 | 1,220 |
| 50 GB (5,000 files) | 2,180 | 2,170 | 2,162 | 2,155 |
| 100 GB (single file) | 2,420 | 2,415 | 2,408 | 2,400 |
Upload throughput is essentially identical across all four storage classes, which is expected since they share the same ingestion infrastructure. The slight variations (under 2%) fall within measurement noise. Parallel composite uploads with gsutil achieved over 2.4 GB/s for large single-file transfers, saturating the instance's network bandwidth.
Download Throughput and First-Byte Latency
Download performance shows more meaningful differentiation between tiers, particularly for first-byte latency on cold requests (objects not recently accessed). We measured both throughput for sustained reads and first-byte latency for individual object fetches.
| Storage Class | Sustained Read (MB/s) | First-Byte Warm (ms) | First-Byte Cold (ms) |
|---|---|---|---|
| Standard | 2,380 | 18 | 22 |
| Nearline | 2,340 | 24 | 48 |
| Coldline | 2,280 | 42 | 120 |
| Archive | 1,860 | 380 | 2,400 |
Standard and Nearline deliver nearly identical sustained read throughput, and their first-byte latency for warm objects is comparable. Coldline shows a noticeable increase in cold first-byte latency (120ms), which aligns with Google's documentation noting that cold-tier objects may require rehydration from deeper storage layers. Archive shows a dramatic increase in cold first-byte latency at 2.4 seconds, making it unsuitable for any workload requiring prompt data access.
Key Finding: For backup and disaster recovery workloads where data is written frequently but read rarely, Nearline offers the best balance of low storage cost and acceptable retrieval performance. The jump from Nearline to Coldline saves only $0.006/GB/month but triples cold-access latency.
Cost-Per-GB Analysis
Storage costs vary significantly across tiers, but the effective cost depends heavily on your access pattern. We modelled three realistic scenarios to calculate the total monthly cost per TB stored, including storage fees, retrieval fees, and operation charges.
| Scenario | Access Pattern | Standard ($/TB) | Nearline ($/TB) | Coldline ($/TB) | Archive ($/TB) |
|---|---|---|---|---|---|
| Hot data | Daily reads, hourly writes | $26.00 | $32.40 | $48.20 | $112.00 |
| Warm data | Weekly reads, daily writes | $26.00 | $12.80 | $14.60 | $68.00 |
| Cold data | Monthly reads, weekly writes | $26.00 | $11.20 | $8.40 | $7.80 |
For hot data, Standard storage is the clear winner because Nearline and Coldline retrieval fees quickly exceed the storage savings. For warm data accessed weekly, Nearline provides a 51% cost reduction over Standard. For truly cold data accessed monthly or less, Coldline and Archive become cost-effective, with Archive offering a 70% saving over Standard for data that is rarely retrieved.
Comparison: GCS vs S3 vs Azure Blob
We ran identical benchmarks on AWS S3 (us-east-1) and Azure Blob Storage (East US) for comparison. GCS Standard delivered 8% higher sustained read throughput than S3 Standard and 14% higher than Azure Hot tier. First-byte latency was comparable across all three for warm objects. GCS Archive access latency (2.4s) is significantly faster than S3 Glacier Deep Archive (12+ hours), making GCS the better choice for archives that may need occasional prompt retrieval.
Pros and Cons
Strengths
- Upload throughput exceeds 2.4 GB/s with parallel composite uploads
- Standard tier delivers the highest sustained read throughput we measured on any object store
- Archive retrieval in seconds, not hours like competitors
- Uniform bucket-level access simplifies permission management
- Object Lifecycle Management automates tier transitions seamlessly
Weaknesses
- Egress pricing remains expensive at $0.12/GB to the internet
- No native client-side encryption integration in gsutil
- Coldline cold-access latency (120ms) is higher than AWS S3 Infrequent Access
- Minimum storage duration charges apply even if objects are deleted early
- Inter-region replication pricing lacks transparent documentation
ByteReview Verdict
Google Cloud Storage earns an 8.7 out of 10. Its Standard tier delivers best-in-class throughput, and the Archive tier's sub-second retrieval time is a significant advantage over AWS Glacier. The main drawback is egress pricing, which remains the most expensive line item for data-intensive workloads. For teams that keep most traffic within GCP, Cloud Storage is the strongest object storage option available in 2026.
