AWS vs Google Cloud vs Aliyun: Same Configuration Performance Measurement and Selection Guide
When cloud computing solutions land, technical teams often face selection problems: in the end to choose the old giants.
AWS
, the network known
Google Cloud Platform (GCP)
, or cost-effective
Alibaba Cloud
?
There is no shortage of articles on the market that are compared by "feature list", but what really determines the cloud experience on the business is often
True performance under the same configuration
. This article will put aside the public relations speech, from.
computing power, network, storage, cost performance
Four dimensions, the three cloud vendors with the same configuration depth of cross-evaluation.
1. test benchmarks and configuration settings
In order to ensure the fairness of the comparison, we selected the three major manufacturers using
Mainstream General Purpose 4vCPU / 16GB Memory
Example as baseline:
Dimension
AWS
Google Cloud (GCP)
Alibaba Cloud (Alibaba Cloud)
Instance Type
c7g.xlarge / m6i.xlarge
c3-standard-4 / n2-standard-4
ecs.g8i.xlarge / ecs.g7.xlarge
Underlying chip
Graviton3 / Intel Xeon
Custom Intel / AMD EPYC
710 / Intel Xeon
cloud disk configuration
EBS (gp3) 3000 IOPS
Hyperdisk / PD-SSD
ESSD AutoPL / PL1
Test node
US East/Asia Pacific (Singapore)
US East/Asia Pacific (Singapore)
Asia Pacific (Hong Kong/Singapore)
Comparison of 2. and Configuration Performance Measured
1. Calculation power and CPU single-core/multi-core performance
In the CPU performance test, the use of Sysbench and Geekbench 6 stress test found that the three focus is very different:
AWS (Graviton3 / Graviton4): Self-developed ARM architecture chips perform well in single-core performance and multi-core throughput. Instance-Graviton instruction execution is extremely efficient and consumes less power and costs when dealing with highly concurrent API services, microservices, and lightweight databases.
Google Cloud (GCP C3 Series): Specializes in high throughput and complex computing. With the processor deeply customized with Intel and the IPU (Infrastructure Processing Unit) developed by Google, GCP has the best single-core peak turbo boost persistence in floating-point operation and deep learning reasoning tasks.
Aliyun (ECS g8i/Yitian 710): It is quite excellent in optimizing common scenarios of Chinese Internet (such as high Web concurrency, Redis cache, MySQL transaction processing). Self-developed ARM
The performance of relying on 710 under multi-threaded concurrent stress testing approaches AWS Graviton3, while on Intel x86 instances, Aliyun's hardware accelerator card (CIPU) significantly reduces virtualization overhead.
Calculation power conclusion: pure CPU computing density and self-developed chip ecology: AWS $\approx $Aliyun> GCP; Deep learning and data computing intensive: GCP $\ge$ AWS> Aliyun.
2. Network throughput and latency (Network I/O)
Network performance directly determines the bottleneck ceiling of a distributed architecture.
[Global Network and Edge Nodes]
AWS: With a large Direct Connect and Transit Gateway ecosystem, connectivity is extremely stable.
GCP: Relying on a global private fiber backbone (Premium Tier) with the lowest cross-border/cross-regional latency.
Ali: Asia-Pacific and mainland China cross-border networks (e. g. GA Global Acceleration) have absolute dominance.
GCP (Google Cloud): After Premium Tier Network is turned on, traffic will enter Google's global private backbone network as early as possible, instead of going through public network routing. This makes GCP's cross-regional (e. g., East to Tokyo) network latency and jitter the best of the three.
AWS: Provides extremely high network throughput limit (SR-IOV and ENA driver blessing), extremely low latency (usually within 0.2ms) between nodes in the same Zone, and is suitable for high-performance computing (HPC) and clustering deployment.
Alibaba Cloud: The node density and line quality in the Asia-Pacific region (especially Hong Kong, Southeast Asia and Mainland China) are unmatched. Its "Global Acceleration (GA)" and "Cloud Enterprise Network (CEN)" are significantly better than those of foreign manufacturers in terms of latency control and offshore dedicated line experience when dealing with offshore and domestic hybrid networking.
3. Cloud disk and storage I/O performance (Block Storage)
Storage is the culprit of many database (MySQL, PostgreSQL) performance bottlenecks. We compared the performance of standard SSD(IOPS limited to 3000~5000 range):
4K random read and write latency (Sysbench FileIO) under the same 3000 IOPS baseline
AWS EBS (gp3) : |████████████| ~1.2 ms
GCP Hyperdisk Balanced: |██████████ | ~1.0 ms
Alibaba Cloud ESSD (PL1) : | ███████ | ~ 0.8 ms
Alibaba Cloud ESSD: Thanks to the CIPU architecture, the storage protocol
Hardening and unloading, the p99 delay performance of Aliyun ESSD is very eye-catching when reading and writing small pieces of 4K randomly, and can be kept within 1ms, which significantly improves the database throughput.
AWS EBS gp3: The biggest advantage is that IOPS is decoupled from capacity. You can only buy a 20GB disk and pay separately to top the IOPS to 16,000. It is extremely flexible and suitable for businesses with write-intensive but small data volume.
GCP Hyperdisk: Dynamic performance tuning is agile and robust in log processing and big analytics scenarios with high throughput (MB/s) requirements.
3. Price vs. Cost Performance (ROI) Analysis
Compare Pay-as-you-go versus annual subscription models without considering Reserved Instances (RIs) or Savings Plans:
Unit price of calculation power (on demand): Aliyun's benchmark pricing in Asia Pacific region is usually lower than AWS and GCP. If Aliyun self-research relies on ARM instance, the cost performance of calculation power will be improved more obviously.
Bandwidth and traffic charges (the largest hidden hole): The public network outbound traffic (Egress Data Transfer) charges of AWS and GCP are relatively expensive (especially cross-zone and public network outbound traffic); Aliyun is more flexible in traffic package and bandwidth billing mode, making it easier for large traffic users to control costs.
Discount Strength: GCP provides ongoing use discounts (Sustained Use Discounts) and reserved commitment discounts (CUDs) to automatically save money without complexity. AWS's Savings Plans have high binding depth but flexibility. Aliyun's discount for long-term package (Subscription) is usually the largest of the three.
4. selection suggestion and scene matching
There is no absolutely perfect cloud, only the solution that best fits the business scenario.
Reasons to choose AWS:
Global business coverage: rapid deployment in Europe, America, South America, the Middle East and other regions.
High ecological dependence: dependence on Terraform, CloudFormation, or need to use rich three-party SaaS plug-ins.
The pursuit of extreme stability: the business is huge and needs to undergo the most stringent industry-level high-availability verification.
Reasons to choose Google Cloud (GCP):
AI/Data-driven applications: Deep use of BigQuery, Kubernetes (GKE), or large model training/inference using a TPU.
Multinational globalization business: attaches great importance to the global user access network low latency and global backbone network quality.
Cloud-native technology stack: The team embraces K8s for a native and seamless hosting experience.
Reasons for choosing Aliyun:
Asia Pacific/China offshore business: core users in East Asia, Southeast Asia or
In China, compliance and high-quality networks need to be balanced.
Cost-sensitive: Hope to significantly reduce infrastructure costs by package or self-developed ARM chips (relying on the sky).
Complex and highly concurrent e-commerce/Web scenarios: The architecture design is close to the domestic mainstream distributed technology stack, and the localization technology support and response speed are faster.
Summary
Under the same configuration,
AWS
wins in the performance of ecological maturity and all-round computing power;
GCP
Unique in specific areas with network backbone and K8s/AI infrastructure; and
Alibaba Cloud
It has demonstrated strong competitiveness in the Asia-Pacific localization network, SSD storage latency and cost-effective computing power.
When selecting the model, it is recommended that the technical team should
User distribution region
and
Core performance bottlenecks (CPU, network, or disk IOPS)
First, establish a small-scale POC test to find the most suitable cloud computing infrastructure for your business.
