Public cloud providers' network performance wildly varies

Top ways to manage cloud complexity
Tonya Hall speak with David Linthicum, chief cloud strategy officer at Deloitte Consulting, to learn exactly what cloud complexity is, the threats surrounding it, and what companies can do to implement best cloud complexity practices.

It is not easy measuring public cloud performance. I know. I’ve benchmarked clouds. So much depends on exactly which cloud configuration you use and your workloads; it’s hard to make objective judgments. But, ThousandEyes, a cloud analysis company, in its second annual Cloud Performance Benchmark, has succeeded in measuring a major performance factor objectively: Public cloud providers’ global network performance.

In this study, ThousandEyes looked at the five major public cloud providers: Alibaba Cloud, Amazon Web Services (AWS), Google Cloud Platform (GCP),  IBM Cloud, and Microsoft Azure. It did by analyzing over 320 million data points from 98 global metro locations over 30 days. This included measuring network performance from within the US using multiple ISPs and global network measurements and by checking out speeds between availability zones (AZ)s and connectivity patterns between the cloud providers. Besides measuring raw speed, the company also looked at latency, jitter, and data loss.  

This is important information. As Archana Kesavan, research author and director of product marketing at ThousandEyes, said in a statement:

“When businesses need to decide which cloud provider best meets their needs, one metric that’s notably missing from their assessments has been performance data, mainly because it’s never been available or has, at best, been myopic. … Understanding cloud performance is essential for planning and for ongoing measurement so you can be assured that you’re providing customers and employees with the best possible performance.”

If you’re looking for an overall winner, dream on. Just like in real estate, it’s all about location. One provider’s great service for your offices in California may be dismal for your location in New York. That said, there are some broad lessons to be taken.  

First, ThousandEyes found some cloud providers rely heavily on the public internet to transport traffic instead of their backbones. This, needless to say, impacts performance predictability. During the evening Netflix internet traffic jam, if your cloud provider relies on the internet, you will see slowdowns in the evening.   

So, while Google Cloud and Azure rely heavily on their private backbone networks to transport their customer traffic, AWS and Alibaba Cloud rely heavily on the public internet for the majority of transport, IBM takes a hybrid approach that varies regionally.

What about AWS Global Accelerator? If you pay for this service, which puts your traffic on the AWS private backbone network, will you always see a better performance? Surprisingly, the answer’s no. AWS doesn’t always out-perform the internet. ThousandEyes found several cases, where the internet performs faster and more reliably than Global Accelerator — or the results were negligible.

For example, ThousandEyes discovered that from your headquarters in Seoul, you’d see a major latency improvement when accessing AWS US-East-1. That’s great. But your office in San Francisco wouldn’t see any improvement, while your group in Bangalore India would see a performance decrease. 

Generally speaking, Latin America and Asia have the highest performance variations across all clouds, whereas, in North America, cloud performance is generally comparable. You need to look at ThousandEye’s detailed findings to pick out the best cloud provider on a per-region basis to ensure optimal performance. Regional performance differences can make a huge impact.

The ISP you use also affects cloud performance. There are significant performance gains and losses depending on which broadband provider you use to connect to each cloud. As always, your network is only as fast as your slowest link. If you’re paying for AWS Global Accelerator, but your offices are cheaping out on slow internet connections, you’ll never see an advantage from using Accelerator.

If you’re moving traffic in or out of China, you’ll pay a performance toll when crossing the Great Firewall of China. Yes, even China-based Alibaba. Specifically, you’ll see comparatively poor performance due to packet loss. ThousandEyes suggests enterprises serving customers in China — but hesitant to pick a hosting region in China due to stringent traffic regulations data privacy — might consider Hong Kong as a viable option. The company found Alibaba suffered the experienced the least packet loss from Hong Kong to China, followed by Azure and IBM.

There is no “best” cloud. There are, however, best clouds for your company and its needs. 

Here are some facts about each provider to consider when cloud shopping:

• First, AWS generally demonstrates low latency and its performance predictability has improved, especially in Asia. But, when compared to Azure and GCP, it still has lower performance predictability due to its extensive reliance on the Internet rather than leveraging its own backbone for delivery.
• Azure continues its strong network performance based on aggressive use of its own backbone to carry user traffic to cloud hosting regions. Despite a slight decrease year over year, Azure continues to lead in performance predictability in Asia when compared to the others.
• Google Cloud continues to favor the use of its backbone for user-to-cloud hosting region traffic delivery and has a strong performance in most regions. But it still has some significant global gaps. For instance, traffic from Europe and Africa takes 2.5- to 3-times longer to get to India, going around the rest of the world instead of taking a direct route. Google Cloud has also decreased visibility into their internal network, making it harder for its users to understand its network paths and performance.  
• Alibaba Cloud offers comparable performance to other cloud providers. It closely resembles AWS in terms of connectivity patterns, regional locations, and even region naming constructs. Also, like AWS, Alibaba Cloud leans heavily on the internet rather than its private network backbone for the majority of user traffic transport. Unlike the other cloud providers, inter-region traffic between Alibaba Cloud regions isn’t contained within its own cloud infrastructure. Instead, it exits Alibaba Cloud, traverses the internet and then makes its way back into Alibaba Cloud.
• IBM Cloud performance is comparable to the others. IBM takes a hybrid approach to traffic delivery. Sometimes IBM uses its private backbone, and sometimes IBM turns to the internet. It appears to be shifting in response to which regions its users are accessing.

“Historically, most IT professionals select cloud providers based on factors such as price or proximity to users, but often overlook the underlying network architecture of the cloud providers, which can have a significant impact on performance,” said Zeus Kerravala, principal analyst, ZK Research in a statement. “The 2019 ThousandEyes Cloud Performance Benchmark Report shows just how widely performance can vary from the top cloud service providers and reinforces the point that businesses should remain diligent about collecting network intelligence and measure their own performance to ensure they are getting what they expect from their cloud providers.”

The message is clear. If you want to make the most of your cloud, you need to look closely at not just price and location, but at overall network performance as well. Only then can you make an informed cloud buying decision. For more on the report, see ThousandEyes’ website. Or, you can try its cloud and website network traffic monitoring services for yourself in a free 15-day trial. 

Related Stories: