What is SKYPE?

Skype chose to cannibalize some of its less essential systems to bring the network up again in the 2010 outage. What factors would go into that choice versus running Skype supernode clusters on Amazon, Google or Rackspace clouds?

• Skype's oflfine IM, group video calling, customer forums, its Skype Manager dashboard for business customers were all taken offline so their servers could be used to make and deploy "mega supernodes." This is a follow-up question to .

• Answer:

  Rob's right on the money when he says "It's hard to answer this on a technical level without understanding the Skype architecture", especially the current state of Skype's infrastructure and protocols. Although the "spin up a bunch of EC2 (or RackSpace :-) ) instances to  compensate for lost supernodes" solution may seem like the obvious choice from the outside, there may be any number of reasons why that was not possible. With absolutely no knowledge of how Skype's internals are assembled, I'm going to start speculating wildly as to potential scenarios: Security concerns - Again, hard to answer without deep knowledge of protocols. (This one is plausible, especially given how keen a lot of governments are on intercepting Skype sessions.) Operating cost - Maybe it would cost ten million USD to spin up the requisite number of brand new instances to fix the issue. (Unlikely, but not impossible.) Bootstrapping brand new supernodes from outside of the network may make things worse. (In certain P2P architectures, spinning up new supernodes ends up eating resources because your existing nodes need to bootstrap its databases, routing tables, etc. If you're not doing this right, you can end up DDoS'ing yourself and creating a second valley in your reliability graph.)