从Azure DTU到常规服务器：已知100DTU，如何匹配RDS的CPU/内存配置？
Got it, let's break this down clearly—converting DTUs to physical server specs isn't 100% precise because DTUs bundle CPU, memory, I/O, and disk throughput into a single metric, but we can get a reliable ballpark based on real-world RDS workloads and Azure's baseline definitions.
Key Context on DTUs
Azure defines 1 DTU as the performance baseline of a small, single-core VM with 1.75GB RAM. But scaling DTUs isn't linear—100 DTUs represents significantly more than just 100x that baseline, especially for CPU and memory (the two most critical resources for RDS).
Recommended Server Specs for 100 DTU Equivalent Performance
Here's what you should target for your RDS server:
- CPU: Aim for 8-12 physical cores (or 16-24 vCPUs with hyper-threading). RDS is often CPU-bound when serving multiple concurrent users, so if you expect heavy user load, lean toward the 12-core/24-vCPU end. This matches the CPU throughput you'd get from a 100 DTU Azure VM (like a Standard D4s v3 or similar, which translates well to on-prem hardware).
- Memory: Plan for 32-64GB of RAM. RDS sessions consume ~1-2GB of RAM per active user, so 32GB handles 15-20 concurrent users comfortably, while 64GB gives you headroom for 30+ users or memory-heavy apps. DTUs include memory allocation, so this range aligns with 100 DTU's memory capacity.
- Storage Bonus: If your RDS workload involves frequent profile loads/saves or app data access, use NVMe SSDs—DTUs include disk throughput, so fast, low-latency storage will ensure you match the I/O performance from your Azure environment.
Quick Testing Tip
Since every RDS workload is unique, start with a mid-tier server (8 cores / 32GB RAM) and monitor CPU, memory, and disk usage during peak hours. If you see consistent high CPU (>70% sustained) or memory pressure, scale up accordingly—physical servers often have less virtualization overhead than Azure VMs, so you might find you need slightly less resources than the estimate.
内容的提问来源于stack exchange，提问作者Leonardo
