Is AWS more reliable than GCP

Submitted by Storage Consortium on November 15, 2019 - 13:02

Munich, Starnberg, Nov. 15, 2019 - New Cloud Performance Report from ThousandEyes shows some significant differences in the network performance of individual providers ...

To the background: Only a few companies that are currently evaluating a cloud provider offer usually have detailed data on network performance. The new (annual) study by ThousandEyes shows the differences in terms of global network performance and connectivity of the five major public cloud providers. Amazon Web Services (AWS), Google Cloud Platform (GCP), Microsoft Azure, Alibaba Cloud and IBM Cloud were examined. The report is thus an aid for IT managers in companies to make qualified decisions regarding investments in the cloud area.

ThousandEyes provides a platform that analyzes how Internet, cloud and other third-party providers influence each other and thus affect the digital user experience. For research purposes, the platform was used to measure network performance in global and within cloud environments. According to the information available, the results come from an analysis of more than 320 million data points collected in 98 global metropolitan regions over the course of 30 days. Comprehensive methods for measuring end-to-end performance and data paths were used. The performance report can thus support companies in obtaining the relevant regional and cloud provider-specific information.

The most important results of the study summarized:

According to the report, some cloud providers rely heavily on the freely accessible Internet for data traffic rather than on their own backbones; a prediction of the performance can be influenced by this. Google Cloud and Azure are increasingly using their private backbone networks to coordinate their customers' data traffic in order to prevent fluctuations in performance that can come with the use of the freely accessible Internet. AWS and Alibaba Cloud use the freely accessible internet for the majority of their data traffic. This creates a higher operational risk that can affect the prediction of performance. IBM pursues a hybrid approach that can vary from region to region.

  • Latin America and Asia therefore have the highest fluctuations in performance across all of the cloud providers examined, while in North America cloud performance is generally constant. Those responsible and decision-makers should select a regionally suitable cloud provider based on the results of the report in order to guarantee optimal performance worldwide. Because regional fluctuations in performance can have a significant impact on performance gains or losses.

  • AWS Global Accelerator does not always perform better than the freely accessible Internet. AWS Global Accelerator was launched in November 2018 to allow customers to use AWS 'private backbone network for a fee. The AWS standard, however, is data traffic over the freely accessible Internet. Despite numerous examples of performance improvements around the world, the Global Accelerator is not a universal solution. Various examples show that the route via the freely accessible Internet can work faster and more reliably than the Global Accelerator. Sometimes the results differ only slightly.

  • Cloud performance is based on the choice of broadband ISP. Companies looking to gain a performance edge should choose their broadband ISP provider based on which cloud they rely on most. This can lead to performance gains and losses, depending on which broadband provider companies use to connect to the respective cloud.

  • The Great Firewall of China causes performance losses for all cloud providers. Although Alibaba has its origins in China, packet loss occurs when passing the Great Firewall of China - just like with all other cloud providers. This shows that Alibaba does not hold a separate position. Companies with customers in China who are reluctant to use a hosting environment in China due to strict data traffic and data protection regulations can consider Hong Kong as an option. Traffic from Hong Kong to China through Alibaba Cloud had the lowest packet loss, followed by Azure and IBM.

Fig. 1: Graphic from the study report / "Cloud connectivity falls into two camps" (source: ThousandEyes Blog post ** on the study).

Results broken down by cloud provider

Last year's Cloud Performance Report compared the three largest cloud providers with the highest market shares, i.e. AWS, Azure and GCP. Alibaba Cloud and IBM Cloud were added this year, so no comparative data is available for these services. There were clear differences between the individual cloud providers in terms of overall performance and connectivity. The most important findings of the research, broken down by cloud provider, are:

  • AWS generally has low latency and overall did better than last year. The metrics for predictability of performance have improved significantly, and there was also a sharp 42 percent decrease in variability in Asia. Compared to Azure and Google Cloud Platform, however, AWS still has less predictability of performance because the provider largely uses the freely accessible internet instead of its own backbone.

  • Azure therefore continues to impress with its very good network performance. The reason for this is the use of the company's own backbone, which is used for user data traffic in cloud hosting environments. Notable changes when comparing 2018 and 2019 data include a 50 percent improvement in performance predictability in Sydney, while India saw a 31 percent decrease in performance predictability. Despite a slight year-over-year decline, Azure continues to lead the way in predicting performance metrics in Asia compared to the other cloud providers.

  • Google Cloud still favors the use of its own backbone for the provision of traffic in the cloud. Users benefit from strong performance in most hosting regions. However, the cloud provider still has some significant gaps worldwide that have not been closed since the report was published last year. Traffic from Europe and Africa takes 2.5 to 3 times longer to get to India as data first travels around the world instead of using a direct connection. Google Cloud also made the internal network less transparent, making it more difficult to analyze network paths and performance.

  • Alibaba Cloud offers a similar performance compared to other providers. The provider is similar to AWS in terms of connection patterns, region locations, and even how regions are named individually. Like AWS, Alibaba Cloud relies heavily on the freely accessible internet for the majority of its data traffic and not on its own private backbone network. What is special here is that the data traffic between regions does not take place between individual branches of Alibaba Cloud, but exits the internal cloud, crosses the Internet and then re-enters the regional cloud network.

  • IBM Cloud is comparable to that of other large cloud providers in terms of performance. IBM takes a hybrid approach to coordinating data traffic, in which the use of its own private backbone and the freely accessible Internet varies depending on which regions the user is accessing.

The conclusion (excerpt) from Archana Kesavan, Director for Product Marketing at ThousandEyes: "Comprehensive information on cloud performance is essential for planning and further action to ensure that customers and employees receive the best possible service." Zeus Kerravala, Principal Analyst, ZK Research: "So far, most IT decision-makers have paid attention to factors such as price or proximity to the user when looking for the right cloud provider. The underlying network architecture has often been completely disregarded, although this has a considerable influence can have on the performance ”.

** ThousandEyes blog post> Key results of the Cloud Performance Benchmark Report

Cross reference: