I visited China this summer with my family. We took the high-speed
train from Beijing to Shanghai, which went 300 kilometers/hour. A few years
ago, I rode a magnetic levitation train in Shanghai with a top speed of 431 kilometers/hour
(or 268 miles/hour). Both rides were a wonderful experience, and even though
the speed was quite high, the rides were very comfortable.
The speed of the trains made me
think of the speed of an Ethernet network. Ethernet speed started with 10 Megabits/second
in the 1980s. Now we see 100 Gigabits/second Ethernet networks being deployed
in the data center on a large scale. It is the exponential growth of global network
traffic driving the speed of Ethernet to higher levels.
Today the human-related network traffic – from tablets, PCs,
digital TVs and smartphones – takes 96 percent of the global Internet traffic,
Virtual Network Index (VNI). Human-related network traffic has been the primary
driver for data center networking growth. So what else could be the force sustaining
the growth of data center networking in the next 10 or even 20 years besides
human-generated traffic? The answer is machine to machine (M2M) communications.
CLICK TO TWEET: How fast is the train to Ethernet speed going in your data center?
defines M2M as direct communication between devices using either wired or
wireless connectivity. Some examples of M2M communications are the monitoring
of fleet vehicle locations, telehealth, unmanned car control, smart homes or
buildings, smart city technologies, etc.
Why would M2M be the primary driver for the long-term growth
of data center networking given that M2M is expected to take only 4 percent of
the global Internet traffic in 2020? First of all, the number of M2M devices (a.k.a.
connected things) will be amazingly huge. By 2025, approximately
80 billion devices will be connected to the Internet, according to IDC.
Second, those connected devices or machines will generate an
enormous amount of data. IDC forecasts that by 2025, the total amount of
digital data will reach 180 zettabytes per year, indicating a 36 percent CAGR (compound
annual growth rate) from 4.4 zettabytes in 2013. For example, a smart home may
generate 1 gigabyte of data a week. Network
World says an autonomous car will churn out 4,000 gigabyte of data per day.
Now, I should point out that a large portion of that 180 zettabytes will be generated
and consumed locally. They will not need further processing or even storage. However,
a good portion of the M2M data will need to be transferred across the internet,
further processed and stored in the cloud. This leads to my third
Processing M2M data will consume compute, storage and
network resources, resulting in increased capacity in the data center. The corresponding
workload may include machine learning, object recognition, speech recognition,
business analysis and medical diagnoses, to name a few.
The data generated by a human being is limited. Each of us can
only consume a limited amount of video time and post limited photos on social
media. However, the data traffic generated by M2M communications will be
virtually unlimited. The further processing and analysis of M2M data will require
scalable compute, storage and networking infrastructure in data centers around