Trade storage for compute. Run efficiently on VPS, bare metal, or modest cloud VMs.
A transparent look at costs, trade-offs, and architecture-based projections.
more storage for projection tables (typical schemas)
more logical ops per vCPU (expected for many workloads)
The math: SSD storage costs ~$0.10/GB/month. A mid-tier compute instance costs ~$70-150/month. Even 4× storage growth (100GB → 400GB = +$30/month) is offset by running smaller servers or fewer instances.
100GB base data, typical SaaS API workload, 90% reads
| Cost Component | Traditional Stack | FraiseQL | Difference |
|---|---|---|---|
|
Compute API servers (AWS m5.large) |
4 instances $280/mo |
1 instance $70/mo |
-$210 |
|
Database PostgreSQL (AWS RDS db.r5.large) |
1 instance $175/mo |
1 instance $175/mo |
$0 |
|
Storage gp3 SSD @ $0.10/GB |
100 GB $10/mo |
400 GB (4×) $40/mo |
+$30 |
|
Cache Layer Redis (AWS ElastiCache cache.r5.large) |
1 instance $110/mo |
Not needed $0/mo |
-$110 |
| TOTAL | $575/mo | $285/mo |
-$290/mo (-50%) |
Prices based on AWS us-east-1 on-demand pricing as of 2024. Reserved instances reduce costs further.
These projections are derived from architectural reasoning, not benchmarks. They represent expected performance for many read-heavy workloads. Validate with your own benchmarks.
| Scenario | QPS/vCPU | vs FraiseQL | Notes |
|---|---|---|---|
| Naive ORM | ~3-5 | ~100-200x fewer | N+1 queries, no optimization |
| Basic Python | ~25-35 | ~15-25x fewer | Single JOIN query, standard serialization |
| Optimized Stack | ~200-300 | ~2x fewer | DataLoaders, Redis, orjson, DTOs |
| FraiseQL | ~500-600 | baseline | Table views, Rust pipeline |
Against naive ORM code, FraiseQL provides massive gains (100x+). Against fully optimized stacks with Redis and DataLoaders, gains are more modest (2x)—but you eliminate Redis complexity and cache invalidation logic.
FraiseQL Cascade returns updated projections directly from mutations, eliminating the need for clients to refetch after writes. This reduces total round-trips and allows higher effective throughput for write-heavy operations.
2-4 vCPU, $10-40/mo
~700-2,200 ops/s
Expected for FraiseQL + Cascade
8 vCPU, $70-150/mo
~3,000-4,500 ops/s
Expected for FraiseQL + Cascade
16+ vCPU
~6,000+ ops/s
Expected for FraiseQL + Cascade
The green shaded area represents cumulative savings. The gap widens as scale increases.
Higher storage cost is far outweighed by compute and cache layer savings.
Read/write ratio >10:1
40-60%
typical savings
Read/write ratio 3:1 to 10:1
10-40%
potential savings
Read/write ratio <3:1
May cost more
sync overhead exceeds benefits
Small Startup
100K requests/day
$1,200
saved per year
Growing SaaS
1M requests/day
$3,400
saved per year
Mid-Market
10M requests/day
$24,000
saved per year
Enterprise
100M requests/day
$180K+
saved per year
Estimates based on AWS on-demand pricing, read-heavy workloads (90% reads), 5× storage multiplier. Your results will vary based on workload characteristics, cloud provider, and commitment discounts.
These are architecture-based projections, not benchmarks. The estimates are derived from understanding how projection tables eliminate JOINs and how the Rust adapter avoids ORM overhead. Validate with your own benchmarks on your specific workload.
The best proof is your own benchmark. Try FraiseQL on your workload and measure the difference.