ZeroFS vs JuiceFS Benchmarks
Performance comparison conducted on Azure D48lds v6 (48 vCPUs, 96 GiB RAM) with Cloudflare R2 backend.
- VM: Azure Standard D48lds v6, West Europe (Zone 1)
- Storage: Cloudflare R2 (S3-compatible)
- Benchmark suite: github.com/Barre/ZeroFS/bench
- Operations per test: 10,000
ZeroFS: Direct S3-only architecture. No additional infrastructure required.
JuiceFS: Requires separate metadata database (SQLite/Redis/TiKV) plus S3 storage. Despite this additional complexity and dedicated metadata layer, JuiceFS performs 175-227x slower than ZeroFS in our tests.
Performance at a Glance
Key Performance Differences
Sequential Writes (higher is better)JuiceFS: 175x slower
Data Modifications (higher is better)JuiceFS: 183x slower
File Append (higher is better)JuiceFS: 227x slower
Empty Files (higher is better)ZeroFS: 1.2x faster
Git Clone (lower is better)ZeroFS: 13x faster
Cargo Build (lower is better)JuiceFS: 22x slower
TAR Extract (lower is better)ZeroFS: 76x faster
Storage Used (lower is better)ZeroFS: 32x less
API Operations (lower is better)ZeroFS: 112x less
Synthetic Benchmarks
| Test | ZeroFS | JuiceFS | Difference |
|---|---|---|---|
| Sequential Writes | |||
| Operations/sec | 984.29 | 5.62 | 175x |
| Mean latency | 1.01ms | 177.76ms | 176x |
| Success rate | 100% | 100% | - |
| Data Modifications | |||
| Operations/sec | 1,098.62 | 5.98 | 183x |
| Mean latency | 0.91ms | 166.25ms | 183x |
| Success rate | 100% | 7.94% | - |
| Single File Append | |||
| Operations/sec | 1,203.56 | 5.29 | 227x |
| Mean latency | 0.83ms | 186.16ms | 224x |
| Success rate | 100% | 2.57% | - |
| Empty Files | |||
| Operations/sec | 1,350.66 | 1,150.57 | 1.17x |
| Mean latency | 0.59ms | 0.83ms | 1.4x |
| Success rate | 100% | 100% | - |
Real-World Operations
| Operation | ZeroFS | JuiceFS | Notes |
|---|---|---|---|
| Git clone | 2.6s | 34.4s | ZeroFS repository |
| Cargo build | 3m 4s | >69m | JuiceFS aborted - no progress |
| tar -xf (ZFS source) | 8.2s | 10m 26s | ZFS 2.3.3 release tarball |
ZeroFS
- Consistent sub-millisecond latencies for file operations
- 100% success rate across all benchmarks
- Completed all real-world tests
JuiceFS
- Failed 92% of data modification operations
- Failed 97% of append operations
- Unable to complete Rust compilation after 69 minutes
- Errors: "No such file or directory (os error 2)" on file operations
Sequential Writes
Creates files in sequence. Tests metadata performance and write throughput.
ZeroFS: 10,000 files in 10.16 seconds
JuiceFS: 10,000 files in 29 minutes 37 seconds
Data Modifications
Random writes to existing files. Tests consistency and caching.
ZeroFS: All operations succeeded
JuiceFS: 9,206 failures out of 10,000 operations
Single File Append
Appends to a single file. Tests lock contention and write ordering.
ZeroFS: All operations succeeded
JuiceFS: 9,743 failures out of 10,000 operations
Empty File Creation
Pure metadata operations without data writes.
ZeroFS: 7.4 seconds total
JuiceFS: 8.7 seconds total
This was JuiceFS's best result, suggesting the bottleneck is in data operations rather than metadata.
Compilation Workload
Rust compilation of ZeroFS codebase. Tests mixed read/write patterns.
ZeroFS: Completed in 3 minutes 4 seconds
JuiceFS: Aborted after 69 minutes with no progress past initial dependencies
Archive Extraction
Extracting ZFS 2.3.3 source tarball. Tests sequential file creation with varying sizes.
ZeroFS: 8.2 seconds
JuiceFS: 10 minutes 26 seconds (76x slower)
Final Bucket Statistics
| Metric | ZeroFS | JuiceFS | Difference |
|---|---|---|---|
| Bucket Size | 7.57 GB | 238.99 GB | 31.6x larger |
| Class A Operations | 6.15k | 359.21k | 58.4x more |
| Class B Operations | 1.84k | 539.3k | 293x more |
JuiceFS consumed 31x more storage and performed 58-293x more S3 operations for the same workload. This translates directly to higher storage costs and API charges.
In our benchmarks, ZeroFS demonstrated 175-227x higher throughput for write operations while using only S3 for storage. JuiceFS, which requires both S3 and a separate metadata database, experienced high failure rates and significantly slower performance across all tests.
The tests also revealed differences in resource consumption: JuiceFS used 31x more storage space and generated up to 293x more S3 API calls for the same workload, which would impact operational costs in production environments.