The Just Works paper
=================
* Another big picture paper
* Written about 20 years after the DP paper
* Now about 10 years old
* What changed between 1988 and 2006? And what's changed since 2006?
* Biggest change between 1988 and 2006 is security.
* Security concerns (viruses, worms, phishing, spyware, spam, DoS attacks).
* Security wasn't a concern in 1988 (one of the first noteworthy worms surfaced
  in Nov 1988).

The key idea in this paper
===================
* The only time something changes on the Internet is when things are broken
* And solutions are deployed just in time
* In other words, if it isn't broken, it will probably not be fixed
* Many examples
--> (1982): DNS, Moving from a single hosts.txt file to a distributed domain name system
--> (1982): Moving from a single level of routing to two levels
--> (mid 1980s): Congestion collapse in the mid 1980s, which resulted in The TCP congestion control algorithm
--> (early 1990s): Policy routing to not carry commercial traffic on the NSF academic backbone
--> (early 1990s): Classless addressing to combat address space exhaustion
* Reason for this just-in-time behavior:
--> It's hard to focus on long-term thinking with exponential growth.
--> Problems won't be solved unless it's critical and imminent

Examples of proposals that were not deployed
===================
* Explicit Congestion Notification (from the late 1990s): Not enough of a need.
  DCTCP (2010) used this within the limited context of datacenters. Again, when
  problems got really out of hand.
* Multicast (both network and application level from the eary 1990s): Again, no
  compelling need at the time. When video got popular post 2010, there were
  other simpler solutions that worked well enough. Like deploying cache servers
  close to user populations---as opposed to multicast from a single server.
* Other examples: Integrated Services, Differentiated Services, Mobile IP
* What's common? These were all good to have, but not critical
* Fear seems to be the main impetus for getting things done.

Worries about the future, and how these have played out since 2006
===================
* Spam, security, DoS attacks. These things seem to have gotten worse with
  time: Heartbleed, government censorship, Mirai and other IoT-based botnets
* Application deployment: E.g., Skype's difficulties getting around NATs.
  Peer-to-peer communication is getting increasingly hard on the Internet
  today. Skype ends up using a relay to work around NATs. But with the
  acquisition by MS, this has gotten easier for Skype.
* Congestion control: Congestion control for high BDP links is quite
  suboptimal. E.g., transcontinental 1 Gbit/s link with a 100 ms RTT. Has led
  to the development of alternatives like Aspera, and more recently BBR (which
  we'll discuss in the course).
* Problems with inter-domain routing: has led to ideas like Internet Routing
  Exchanges. Partially fixed by approaches like SDN.
* Mobility especially related to roaming in SSH: Has led to approaches like
  mosh for SSH while roaming.
* Address space depletion: IPv6 is finally seeing adoption now that we are
  actually running out of addresses (cf. IPv6 Launch Day in 2012).
* Since the paper was published many networked applications have become
  popular: social networks, IM apps, YouTube, Spotify, etc. But almost all of
  them use the client-server model and run on top of HTTP, which doesn't have
  problems with NATs. HTTP turns out to be good enough.
* Overall, the paper seems to have been prophetic.

My take on the paper
===================
* Problems do seem to get solved in the nick of time; otherwise the steady
  state seems to be stagnation.
* But that seems to be an unavoidable consequence of a large infrastructure
  with multiple stakeholders. In fact, I don't know of any other large
  infrastructure that actually _changes_ as frequently as the Internet does.
  For instance, when was the last time that the power grid or the telephone
  system changed? In a way, the Internet has become a critical part of the
  infrastructure, and the fact that it changes even occasionally is quite
  remarkable.
* Also, plenty of change has been happening at the application layer (FB,
  WhatsApp, SnapChat, Spotify, Netflix, etc.).
* Changes at the application layer (which provides plenty of knobs) are good
  enough until people get deeply frustrated.
* This frustration then leads to changes at lower layers (e.g., SDN,
  programmable routers, BBR, QUIC, DCTCP).

Summary of both papers
===================
* The Internet is a best-effort network that puts connectivity above all else.
* It's mostly in a state of good enough until it gets to a point of really bad,
  and then it reinvents portions of itself.
* The great news is that reinvention still continues to happen.
* SDN, QUIC, BBR, programmable networks, datacenters are < 10 years old.