Traffic Jam?

This week, the first week in which school was out of session for the summer, I noticed that my commute to work was much shorter than it had been, reduced from about 25 minutes to 15 minutes. It’s always hard for me to believe that such a simple thing, as fewer drivers on the road due to summer vacations, is enough to cause such wild swings in commute times. I took advantage of the additional time that I’ve gained, and read the Cisco Visual Networking Index (VNI). Then I began to think about how datacenters and networking infrastructure handle their increases in traffic.

The VNI contains a huge amount of data, focused on global IP traffic. In 2014 there was a 500% increase in IP traffic compared with 2009, and IP traffic is expected to show double-digit growth (23% CAGR). Sprinkled throughout the Cisco document is the primary reason for this growth – video. If a picture is worth 1,000 words, how many words is a video worth?

First, in terms of data requirements, we need to set the record straight—1,000 words written into a simple text document should use about 1KB of space. A typical digital picture typically uses 1 to 2MB of data. It appears to me that a picture is worth more like 1 million words! For additional comparisons, a one-minute audio file also takes about 1MB of storage, while one-minute of video utilizes about 10MB (more if it is HD quality).

With these simple comparisons, it is easy to see that video streaming, whether it’s from YouTube or Netflix, is a large contributor to the increase in IP traffic. The VNI report confirms this, stating that consumer Internet video traffic represented 64% of all IP traffic, and is forecasted to grow to 80% of traffic by 2019. That’s a lot of video! And it gives us a good idea about why datacenters and networking infrastructure seem to be constantly growing.

Let’s quickly walk through the thought process. More IP traffic requires more server bandwidth. Server bandwidth increases are met by a combination of the growth in the number of servers as well as increased bandwidth capacity per server, as evidenced by increasing CPU cores, faster links, and more memory per server. In addition to the need for more infrastructure equipment to handle the increased requirements, it is also important to replace old servers with new units, ensuring that latest technologies are being used.

I have mentioned before in other blogs that the memory industry is doing its part to help alleviate some of the bandwidth issues by driving the upgrade cycle to DDR4 technologies that allows for bandwidth increases of ~10% per year. In some more extreme cases, we see at least some SoC/ASIC companies looking to add a real “pop” to their bandwidth by shifting toward even higher-bandwidth memories like HBM.

It seems the entire tech industry, from datacenter and networking infrastructure developers to semiconductor designers, are innovating in response to existing challenges around big data that will help prevent an IP traffic jam. Rest assured, the industry is looking forward to rising to this challenge.