Azure、AWS、GCP（クラウドパフォーマンス）、ThousandEyes

ThousandEyeのクラウドパフォーマンスに関する最初のレポートでは、Azureが最も高いパフォーマンスの安定性を示し、AWSは最も低いパフォーマンスを示しました。Microsoft’s Azure cloud platform trounced Amazon Web Services (AWS) in performance stability, according to a new report from ThousandEyes. That report compared the global performance of the three major public cloud providers — AWS, Azure, and Google Cloud Platform (GCP). The ThousandEyes report, “Public Cloud Performance Benchmark Report,” showed that while the three providers demonstrated generally consistent performance, geographic performance variations exist between the providers, particularly in Latin America and Asia. Overall, Azure demonstrated the highest amount of performance stability, and AWS the least. “We know all clouds were not created equal but we have data now to prove that there are architecture and connectivity differences between AWS on one side versus Google and Azure on the other side,” said the report’s author and ThousandEyes Senior Product Marketing Manager Archana Kesevan. ThousandEyes set out to fill a gap in research comparing the big three providers in their “race-to-cloud domination” said Kesevan. “There’s a lot of deficit in terms of performance data and we realized that that is very important factor when it comes to understanding the cloud and making these cloud decisions.” The report collected data from the ThousandEyes Network Intelligence platform. The platform relies on monitoring agents that are pre-deployed across global locations within the three cloud providers environments and the internet to track the performance between cloud regions. For the analysis, ThousandEyes collected data on network latency, loss, jitter, and network path. In order to collect performance data over time, it mapped four measurements over 30 day periods: end-user data from cloud regions; inter-availability zone and inter-region within the same cloud provider; and performance between the three providers for multi-cloud. The result was 160 million unique data points on the performance of these cloud providers. The report found a number of regional anomalies in performance between AWS, Azure, and GCP. The first being that AWS had less performance stability in Asia. This is primarily because AWS relies on the internet more than GCP and Azure, which route traffic on their own backbone infrastructures. Routing over the internet, as an unregulated network, makes it more vulnerable to attacks and outages. Kesevan said that in Asia particularly AWS’ reliance on the internet led to large variances in network latency. In this region AWS demonstrated 35 percent less network performance ability than GCP and was 56 percent slower than Azure. Google, on the other hand, has three-times higher network latency from Europe to India. The reason, Kesevan said, is that while Google’s backbone is well-connected and has great service level agreements (SLAs), it doesn’t have a direct route from Europe to India. “If you’re an enterprise looking to expand into Asia and you’re considering Google Cloud, you’re not going to get the most optimal performance from Google.” The third regional finding showed that when connecting Europe to Singapore, AWS and GCP were 1.5 times slower than Azure. While the report did highlight certain performance discrepancies and variations between the three cloud providers, choosing a clear “winner” was harder than expected. “Overall, I think Microsoft, from backbone and the way they’re approaching this, in my eyes would put them on top. The caveat being that every region is different and every use case for a cloud is different,” said Kesevan. She explained that Google does gives its enterprise customers choices via pricing tiers when it comes to the cloud and the backbone, which can improve end-user experience, but that AWS overall has the performance variations because they end up making cloud decisions for you. At the end the report demonstrated that — aside from regional anomalies — performance-wise the three providers were all similar. Instead of selecting one cloud provider, enterprises might lean toward multi-cloud deployments. “The data does show something, which is to not assume uniform performance across geographic regions,” said Kesevan. In the end, multi-cloud might be the true winner.