THRON Content Delivery Network performances

THRON Content Delivery Network performances

There are things which can not wait, and one of those is watching videos: in fact, whenever a user wants to watch a video, he surely does not want to wait for loading time. How can we understand how long do our customers need to wait before they can start watching a video? We want to understand THRON delivery performance for real users around the world.

 

How THRON delivers content

THRON is designed to create, store, manage and use digital content. In order to achieve optimal performance and reliability, it leverages distributed architecture over a cloud computing infrastructure, providing several “nodes”. Each one of these nodes is capable of managing user requests such as watching a multimedia content.

A Content Delivery Network (CDN) is a network providing nodes, often referred to as "Points of Presence" (POP), which cache content remotely and allow content to be delivered by those nodes which are closer to the end user, avoiding packets to travel all the way to the origin servers.

THRON delivery is enhanced using an integrated CDN, its acceleration is being used both for multimedia content (like video/audio) and static content (like documents/images).

Let's consider the toughest scenario in terms of amount of data to be transferred, hence potential waiting time for end-users: this happens when video content are delivered via progressive download in HD quality. In this scenario video quality (bit rate) doesn't follow user bandwidth capacity and the risk of long buffering time is at its highest. 

This article explains how measurements have been made and provides analysis results.

 

Measurement setup

Performance measurements can be achieved in many ways, most of them focus on analyzing maximum server throughput by measuring performance within the same node/cluster/network or by using powerful internet data centers. Those measurements are usually applied to identify maximum theoretical performance, ignoring real user infrastructure because it creates a bottleneck.

A different method to measure performance is to focus on what real users will experience on their devices when visiting websites/projects, obtaining a Real User Monitoring (RUM). In order to obtain RUM, we use a standard browser (Google Chrome) that runs in a Catchpoint (www.catchpoint.com) node. A Catchpoint node is a standard PC connected to a standard connection, opening a specific web page.

We don't need any special plugin to obtain data because we exploit common javascript events in order to understand:

  • when the page has been loaded,
  • when the player has started playback,
  • when the player has ended playback.

Since some javascript efforts (loading times) are necessary in order to obtain those data, empiric tests demonstrate that we can consider 500 ms of overload while evaluating the total amount of time spent before video playback and after its conclusion. 

For this test we are going to use the "Big Buck Bunny" video that can be used without restriction (https://peach.blender.org/about/) with a duration of 26.8 seconds. THRON player uses the HD version of the content which has a bitrate of 2 Mbps and the Progressive Download HTML5 Technology.

Each test has been grouped by geographical region and city.

To provide a readable result set, only average values are provided. Values variance (deviation from average) was very low anyway.

 

Results measurement

Results are provided in the following table after a brief description of table’s columns. 

City: the city from which the measurement has been made.

TTPB: Time to playback, it defines the delay time from the load of the webpage to the player "playback started" event. The webpage only contains a THRON Player embed code. Browser needs network time to download resources and some other time to run the javascript code and "render" the player. When the player launches the "playback started" event, the playback is started and the user is watching the video. The lower this value, earlier the user can start watching the video.

TTC: Time To Complete, defines the delay time from the "playback started" event to the "playback ends" event. The video duration is 26.837 ms. The closer this value is to the video duration, more the playback of the content is smooth.

% Delay: The ratio between time spent for watching the video and the effective content duration. Empiric tests show that delay lower than 4% are not perceptible by human viewing the content so it means no interruptions.

 

 

City TTPB [ms] TTC [ms] % Delay
Beijing - China 2,500 27,988 2.33
Berlin - Germany 830 27,149 0.00
Bloomfield - New Jersey 1,436 27,521 0.79
Buenos Aires - Argentina 1,719 27,022 0.00
Hangzhou - China 2,019 28,039 2.47
London - UK 1,014 27,607 0.98
Los Angeles - California 1,492 27,543 0.92
Madrid - Spain 1,015 27,030 0.00
Melbourne - Australia 2,921 27,108 0.00
Moscow - Russia 1,142 27,238 0.11
Paris - France 969 27,312 0.14
Quingdao - China 2,084 28,273 3.31
Redmond - Washington 2,120 27,148 0.05
Rio De Janeiro - Brasil 2,563 27,550 0.77
Riyadh - Saudi Arabia 1,676 27,411 0.29
Rome - Italy 956 27,231 0.29
Seattle - Australia 2,113 27,418 0.74
Singapore 2,188 27,079 0.00
Stockholm - Sweden 973 27,472 0.54
Tokio - Japan 1,630 27,076 0.00
Vancouver - Canada 1,791 27,033 0.00

 

 

 

Quality

HD video quality in progressive download is used on all tests.

 

Video playback interruptions

No video playback interruptions have been recorded during all tests, not even from China. Progressive download with HD quality represents the toughest scenario for the end-user, because bit rate is not adaptive. We will create another test scenario for a better understanding of the quality used by end-users with adaptive bitrate.

 

Conclusions

Using THRON CDN provides best performances for worldwide content distribution with HD quality.

Videos playback is the most complex delivery case, so we can assume every content delivery (documents, images, audios) will achieve same or better results.

Was this article helpful?
3 out of 3 found this helpful

Have more questions?

SUBMIT A REQUEST

Hai altre domande?

INOLTRA UNA RICHIESTA

Comments

  • Avatar
    Antonio Comelli

    We have updated this article, introducing some results of performance measurements made in different nodes located in many parts of the world.