Ross Johnson (moderator)
Andrew Blum (Journalist)
Mary Meduski - President of Tierpoint Compass
Sharif Fotouh - Founder, CEO Edgepoint
Four topics of discussion:
- The physical components of networks
- Current issues
- Future issues
- Personal stories
Physical components:
Andrew Blum, as part of an article for Wired magazine, decided to physically follow the cable that came to his home back to the network, leading him around the world. He found that the internet hubs connect at fewer places than one may think - there are 12 buildings connecting around 300 separate networks, and in those buildings there is far more connectivity of networks than in any other physical connection point. This almost guarantees that any web page downloaded or email sent goes through one of those 12 buildings. Similarly, there are about 12 major cables going across the ocean connecting continents.
One of the biggest challenges is addressing the massive amount of traffic. In 2021 there is projected to be 127 times the traffic on the internet than was in 2005. Most of this internet is for residential usage, and most of that is for video. In 2016, 80% of traffic was video, and 50% of that was on smartphones. The smartphone revolution in 2007 caused expectations to be able to see anything anywhere, and it skewed network activity towards download, mostly for video. Streaming video companies have also contributed to the trend on the supply side: Netflix, for example, has more subscribers in the US than the combined total of all cable providers. It also accounts for 40% of peak traffic on the internet. Social media is also very video centric.
This has driven the network to optimize towards video delivery: Introduction of CDNs bring content closer to consumers, and the CSPs invested 1.6 trillion in infrastructure in the last 10 years to help provide more local connectivity, bypassing the big pipes.
Issues we are facing today:
Infrastructure optimization to one task: So over all, statistics show we fixed the video problem (except in the rual/underserved areas). However, now we have an internet optimized to handle one type of content going in one direction 80% of the time. This type of infrastructure is going to be challenged as we move into the world of IOT, where there is a lot more upload. For example, video security cameras uploading video, hundreds of millions of sensors of all type uploading data. The infrastructure is optimized for download, not upload. In most residential internet connections there isn't symmetrical upload/download capacity, and this will continue without adoption of technology like fiber.
The big question is whether all this uploaded data needs to be uploaded all the way to central hubs? Some data surely can remain local. For example, why would weather information for Boise, Idaho need to be uploaded to the same server as weather for Beijing, China - or even San Francisco, California. Local weather information is consumed locally, and can be kept mostly locally. So doing more locally (compute and network) at the edge could mitigate the problem. Today only 10% of all compute power is done at the edge; in the future that will be changing to 50%. This reverses a previous trend of pushing everything into the cloud - now we will be moving back to the edge. The small data centers at the edge can accommodate more compute power and network activity, to offset work done deeper in the internet at larger data centers.
Other issues include
- availability of broadband. In the US, only 50% of low income families have broadband. Would we accept such a statistic for water or electricity?
- The internet giants are building their own internet, laying out their own infrastructure and generally taking care of their own needs separately. Internet is rapidly being replaced by "Google-net", "Facebook-net" or "Amazon-net". A single company owning the end to end internet layout is a major concern for the concept of a free internet. Need to make sure there will be openness in the infra running the internet.
No comments:
Post a Comment