New Podcast
Managing Traffic During Peak Demand; Plus, Microsoft, Akamai Outages

Internet and Cloud Research

AWS Addresses Performance Instability with AWS Global Accelerator

By Archana Kesavan
| | 7 min read

Summary


Yesterday, at re:Invent 2018, Amazon made a plethora of announcements ranging from a new base-station as a service (AWS Ground Station) to introducing re:Inforce, a new security conference. But one announcement stood out in particular—the unveiling of AWS Global Accelerator—their plan to allow AWS customers to ride their private backbone… for a price.

With this announcement, AWS joins a host of other public cloud providers that also are monetizing their backbones—an interesting new trend for those keeping track of key differentiators between the various providers. Also of note, this announcement comes just three weeks after we unveiled the findings of our first-ever Public Cloud Performance Benchmark Report, a comparative study that measured network performance between AWS, Azure and GCP, which revealed some significant AWS performance anomalies and global variations. Turns out, the absence of such an offering like AWS Global Accelerator was exactly the culprit. Let's dig more into that here.

The Internet - The X-Factor of Public Cloud Performance

Our recent research that compared network performance and connectivity architecture of the three largest public cloud providers—AWS, Azure and GCP—unveiled very interesting trends in global performance. One of the key highlights from the benchmark report calibrated AWS to have the least performance predictability, especially in Asia. We noticed that AWS deployments had the largest standard deviation in network latency, as seen in Figure 1 below.

Bi-directional latency variations in Asia
Figure 1: Bi-directional latency variations in Asia reveals AWS to have the largest variation in network performance and thereby least performance predictability.

Upon further investigation, we got our first clue as to why such an anomaly would exist, especially in one of the fastest growing markets in the world—Asia. Analysis of network path data revealed that AWS deployments are subject to higher reliance on the Internet thereby resulting in greater performance variation and operational risk.

AWS’ network design forces user traffic to stay longer on the public Internet and only enters the AWS backbone closest to the target region. This behavior is commonly known as hot-potato routing and is a symptom of AWS not anycasting public routes associated with each of their regions from global edge locations, resulting in traffic always flowing across the Internet to their regions. This means if your service is hosted in AWS us-east-1 (AWS’ famous Ashburn data center), then traffic from your users, irrespective of their location (as seen in Figure 2), will enter AWS’ backbone closer to Ashburn. Traffic stays longer on the public Internet, a best-effort network, rather than riding AWS’ well connected, congestion-free backbone. This behavior is a stark difference when compared to how Microsoft Azure handles traffic, but that's a discussion for another day. If you are curious though, I would urge you to take a look at the 2018 Public Cloud Performance Benchmark Report to understand the difference.

AWS connectivity architecture without Global Accelerator
Figure 2: AWS connectivity architecture without Global Accelerator.

Why Did AWS Launch the Global Accelerator?

The AWS Global Accelerator is a solution to alter this behavior. Instead of riding the Internet to your AWS hosting regions, you now have the option to push traffic into the AWS backbone closest to the end-user and ride the AWS network to the hosting region. What does that mean for enterprises? Well, a) you get to go over a network that is more reliable than the public Internet, therefore better performance and b) you are going to fork over some $$ to AWS for this luxury. Well, there is no such thing as a free lunch.

Trend Alert - Monetization of the Cloud Provider Backbone

The AWS Global Accelerator service offering did not really surprise us. It is a way for them to monetize their internal infrastructure and that seems like a natural progression. You can pay for performance, or you can save on costs and sacrifice user-experience. With this announcement, AWS joins Google Cloud, who also offer a very similar service through Network Tiers. AWS has been in the news recently for their investments in undersea cables and is constantly building out their internal infrastructure. So it doesn’t come as a shock when they introduce a service that lets them monetize that infrastructure. Turns out Azure is the only one who hasn’t joined the backbone monetization party, but only time will tell if that's to change or not.

Ultimately, whether your business-critical traffic is riding the Internet or a cloud provider’s network, it’s not in your direct control. That’s why so many SaaS companies and Global 2000 enterprises rely on ThousandEyes to monitor app experience and networks beyond their four walls. To learn more and stay tuned on ThousandEyes, subscribe to the ThousandEyes Network Intelligence blog.

Subscribe to the ThousandEyes Blog

Stay connected with blog updates and outage reports delivered while they're still fresh.

Upgrade your browser to view our website properly.

Please download the latest version of Chrome, Firefox or Microsoft Edge.

More detail