It has been a day from hell for Skype fans and Skype, the company. The outages have impacted many. Skype’ s misfortune turned out to be a boon for SIPphone. The company saw a 400% increase in traffic this morning,…

http://digg.com/software/Skype_Aftermath_Technical_analysis

While I am monitoring the Skype network “heal” itself the results are not near perfect. The “Distributed Data Store(see DHT)” is way behind. I am still receiving messages that are 24 hours late and my contact list(s) are not in “sync”. This so called “resilient network” is starting to make shape. If you are not “in the know” then please see my other writing about this outage. Here is what really happened:

•   Skype employees introduced code into the "login/connectivty" server farm that was not compatible with current Skype clients "see Morpheus getting booted from (fasttrack).
•   This is a single point of failure even in the masses of 9,000,00 concurrent users.
•   This is a single point of failure for ~220,000,000 users.

Question:
• Why did resolution take so long and is still ongoing?
•
Answer:
• Skype uses a Peer to Peer topology. This consists of Supernodes which maintain a “DHT” type layer between other Supernodes. This data is routed down to their Children (300 – 500 at any one time).
• The “DHT Type” layer is responsible for presence, avatars(icons) and above all “Call Routing”.
• Child nodes do not know of this such layer thus they depend on Supernodes.
• If a Supernode goes offline then all Children are cut off until they find another Supernode.
Avoidance:
• If child nodes knew about the upper layer at any point then this would be more resilient to outage because they are not dependent on Supernodes.
• Local Discovery: If Skype had a layer that made local availabilty to all users then in an office environment Skypers would be able to locate each other without the need of a Supernode. (“this excludes Bonjour as it does not relay local node cache”).
• Login failures.. This is pure redundancy(“central servers”) however Skype had this in place, but it failed because they input flawed code that was not assuring complete SSL based authentication.
• Skype Employees need to come clean and stop blaming this on “very old” bugs, test before you release, surely there is a test bed???
End:
• Skype is currently re-creating it’s decentralized network from scratch, from the end user to the Supernode users to the entire “distributed index”. This is done over UDP and takes some time so do not trust the following until the network is again stable:
• Prescence
• Profiles, this includes avatars.
• Contact list, yes you may see people that you blocked, give it time to heal.
• Stability, connections may continue to drop because your Supernode went offline and your local cache is no “doing it’s job”.

Summary:
• Decentrilzed networks like Skype are subject to global outage the same as Kazaa and possibly Joost.

Take Care

~Julian Cain

jolix (at) mac (dot) com

If I remember this was also a Issue when Sharman locked unofficial clients out a few years back .