While many think of Hong Kong as just another part of China, it has a separate designation as a “Special Administrative Region” of the PRC, making it a semi-autonomous entity on the global stage. This distinction isn’t just political; it carries over to internet connections as well.
As such, we have been preaching that in order to deliver a great user experience inside Mainland China, you must deliver from inside the mainland itself, rather than from Hong Kong or other APAC locations.
We have been investing heavily in our node infrastructure in China, with 15 nodes on the mainland (plus an extra three in Hong Kong), and 15 more coming in a few days. These nodes rely on a Command and Control (C2) system based in both Hong Kong and Singapore. But with our node growth, we have noticed some performance and availability issues to reach that C2 system.
So we decided to investigate the build-out of our new C2 for our nodes inside China, and in Beijing in particular. The systems came online this week and we started testing the performance…the results are just staggering!
Here are two side-by-side views of the test going to HK vs. the test to China (the top graph shows mean time, and the second shows median).
As you can see, all of the key metrics went down: connection time, first byte received, and response time!
Here is another view by city:
Not only the performance improved, but the reliability as well, no more spikes and weird performance behavior we used to see when going out of China during peak hours.
In this chart, which shows response time based on the hour of the day (China Standard Time), we see that both metrics experience the type of spikes that would be expected when more people are accessing the internet, but those connections within the mainland (the blue bars), are far more stable than those between the mainland and Hong Kong.
The results yield a simple and definitive conclusion: delivering great performance inside China requires having a delivery and DNS footprint inside.
Mehdi – Catchpoint