Skype's voice over IP network
=============================

Overview
--------
 - services to find other users, voice and video conference,
   media sharing, etc.

 - more than 600million users, $600billion per year,
   accounting for more than 1/8 of all international call minutes,
   typically over a hundred thousand calls active at any time

 - uses JoltID's global index system to find anyone who has been
   logged into Skype in the last 3 days.  It is generally assumed
   that JoltID's distributed query system to handle this is based
   on a form of sharding.

Approach to scalability
-----------------------
 - Skype maintains one central account system to track login ids
   and passwords, as well as a small number of default servers,
   but the vast majority of the system is a decentralized network
   run using client machine resources
 
 - Each client machine (with some limitations discussed below) might
   be used as a 'supernode': a small hosting or server node within
   the Skype network to carry out the global index system lookups
   (i.e. to host a shard) and to coordinate or forward call information
   for other clients.

Picking clients to act as supernodes
------------------------------------
 - generally a client will not be chosen to host a supernode if
   they are behind NAT, a firewall, or proxy server, nor will they
   be chosen if they are particularly bandwidth or cpu-limited

Bypassing firewalls etc
-----------------------
 - Skype tries to make the process of placing calls as technically
   simple as possible for the user, and thus takes on responsibility
   for finding its own way past/through any firewalls, proxy servers,
   NAT, etc that might exist between the people taking part in a call

 - because the call traffic Skype generates is precisely the kind of
   traffic many systems seek to block or limit, it must take a variety
   of steps to get past/through those

 - on install, Skype picks a random port to use (hence firewalls etc
   can't block skype for all users simply by blocking a single known
   port)

 - the Skype client maintains (e.g. in the registry on Windows) a list 
   of supernodes it knows about and has been able to access recently.

   On install this list is populated with a small set of servers run by
   Skype, but after that the list is populated with supernodes running
   on different client machines.

   On startup it goes through this list trying to access any one of those
   nodes to connect to the rest of the Skype network, check for updates,
   etc.  This list is periodically rebuilt/refreshed, and if no supernodes
   can be accessed the client is unable to proceed.

 - on startup, skype also tries to assess what kind of obstructions there
   might be, and how best to bypass them, by trying a number of 
   different connection ports and protocols until it finds something
   that gets through (e.g. it will try tcp and udp on its random port,
   on port 80 acting like http traffic, on port 443 acting like https,
   etc)

 - if necessary, Skype tries to masquerade as a flavour of 'accepted'
   traffic, setting up a connection in that form with a supernode,
   and getting that supernode to carry out searches and connection
   setup with other nodes to establish a call.

Privacy and other concerns
--------------------------
 - Skype claims that the added bandwidth and cpu usage is 'negligible'
   for chosen clients, but there is contention from system administrators
   and other users

 - mistrust of Skype is aggravated by the way the skype application tries
   to mask itself in order to bypass firewalls and other 'obstructions',
   often causing administrators to liken Skype to a virus

 - there are also privacy concerns given that call and lookup information
   is being stored, albeit temporarily, on other peoples' machines