Performance vs Stability: Comparative Evaluation of Major Public Cloud Providers

Since the rise of cloud computing, many vendors boast impressive performance benchmarks, but for enterprises the real concern is long‑term stability rather than short‑term speed.

The author argues that performance figures are often unreliable, unnecessary for most business workloads, and not durable over time, emphasizing that stability should be the top priority when selecting a cloud service.

To assess stability, the author purchased identical cloud instances (4 CPU / 8 GB RAM), databases (≈6 GB memory) and storage services from AWS, Azure, Alibaba Cloud, Tencent Cloud and UCloud, installed monitoring tools, and recorded performance 24 hours a day for seven days.

Stability was quantified using the coefficient of variation (CV), where a larger CV indicates poorer stability.

CPU stability: All three testing methods (Sysbench prime calculation, π‑calculation, Ubench) produced consistent results. AWS showed the best stability, followed by Alibaba Cloud, UCloud and Azure, while Tencent Cloud displayed the most fluctuation.

Memory stability: Tested with mbw and Ubench; AWS and Alibaba Cloud ranked first, Azure and Tencent Cloud followed, and UCloud lagged significantly.

Disk stability: Using the industry‑standard FIO tool, Azure, Alibaba Cloud and Tencent Cloud showed the best read/write bandwidth stability, while AWS and UCloud performed worse.

Database stability: Using MySQL and the Sysbench QPS metric, Alibaba Cloud outperformed all others by several orders of magnitude, while Azure China showed the poorest stability.

Storage stability: Using Apache JMeter to issue get/put/delete/head requests via each provider’s Java SDK, AWS and Azure China ranked highest for GET stability, Alibaba Cloud was slightly lower, and the popular storage services Qiniu and Kingsoft Cloud performed the worst.

Overall, AWS and Alibaba Cloud form the leading stability tier, with AWS excelling in CPU, memory and storage, while Alibaba Cloud matches AWS on most metrics and leads in database stability; Azure occupies the second tier, and other Chinese providers such as Tencent Cloud and UCloud show mixed results.

Factors affecting stability include CPU turbo boost, NUMA memory configuration, IOPS throttling by providers, and resource contention on shared physical hosts, especially as user counts grow.

The author concludes that the era of pure performance competition is ending, and the ability to deliver stable services at massive scale will become the decisive advantage for cloud providers.