Report: AWS Performance Predictability Related to Internet Reliance ThousandEyes published cloud performance research which cites AWS’s dependence on the public Internet, rather than its own backbone network, as contributing to operational risks that can affect performance predictability. The San Franciso-based firm, which calls itself an Internet and cloud intelligence business, published its second annual Cloud Performance Benchmark this month. It compares global network performance as well as connectivity differences between five major public cloud providers: Amazon Web Services, Google Cloud Platform (GCP), Microsoft Azure (Microsoft Azure), Alibaba Cloud, and IBM Cloud. [Click on the image to see a larger view.] Cloud Connectivity falls into two camps. (source:ThousandEyes). The company stated in a press release that some cloud providers rely heavily upon the public Internet to transport traffic rather than their backbones. This can impact performance predictability. AWS and Alibaba Cloud, while Google Cloud and Azure heavily rely on their private backbone networks for transporting customer traffic, protecting it against performance variations associated with delivering over public Internet, rely heavily upon the public internet for most transport. This can lead to greater operational risk and impact on performance predictability. IBM uses a hybrid approach that is specific to each region. In a blog post, the company explained this finding and advised customers to pay attention. It stated that although the cloud providers tested generally had comparable performance in terms of bi-directional network latency and architecture, differences in connectivity can have an impact on traffic between users or certain cloud hosting regions. “For example, Azure and GCP use their backbones extensively to carry user-to-hosting-region traffic. AWS and Alibaba rely heavily on the Internet to transport user traffic, while IBM uses a hybrid approach. Exposure to the Internet increases uncertainty in performance, creates risk to cloud strategies, and raises operational complexity. Enterprises planning public cloud connectivity should consider the tolerance of their organization for the unpredictable nature. The report also questioned the AWS Global Accelerator, described by AWS as “a service that improves availability and performance of applications with local and global users.” AWS stated that it provides static IP addresses as a fixed entry point to AWS Regions application endpoints. ThousandEyes stated that “AWS Global Accelerator does not always outperform the Internet.” The Global Accelerator follows an optimized route through AWS’ densely connected backbone. However, performance improvements are not uniform across the globe. While the Global Accelerator is more performant than the Internet connectivity path in many cases, there are still instances of performance improvements that are negligible or worse than default AWS connectivity. Enterprises should evaluate the performance gains of the Global Accelerator before deciding on a strategy to maximize their ROI. ThousandEyes also listed other findings in the report:

  • Cloud performance anomalies can be significant depending on the provider, the hosting region and the user location.
  • All cloud providers, including Alibaba pay a performance fee when they cross the Great Firewall of China.
  • Cloud performance is affected by the choice of US broadband ISPs.

The company commented further on the AWS cloud and stated that “The major public cloud providers (hopefully) are continuously optimizing their networks to (hopefully] improve performance and stability for customers.”