[GUFSC] BBC News - Skype crash: Software bug and server overloads blamed
Altamir Dias
altamir em emc.ufsc.br
Quinta Dezembro 30 17:02:28 BRST 2010
Skype crash: Software bug and server overloads blamed
Server overloads and a bug in Skype for Windows caused the two-day
outage for the net phone firm.
Details of what caused the service to be unusable for millions of users
prior to Christmas have been posted on the firm's blog.
The two events combined to create a cascade of problems that managed to
knock out much of the network underpinning the phone service.
Skype is assessing how its network is built to stop the problem recurring.
Traffic cascade
Writing on the Skype blog, Lars Rabbe, chief information officer at the
company, said the problems started on 22 December, when some of its
servers that handle instant messaging started getting overloaded.
This meant that the responses they sent to Windows machines running
Skype were slightly delayed. Unfortunately, a bug in one version of
Skype for Windows meant this delay caused the program to crash.
About 50% of all Skype users ran the buggy 5.0.0.152 version of the
software, said Mr Rabbe.
This caused problems for Skype because of the way the network supporting
it is organised. Some of the data travelling round Skype's network are
passed through all those machines logged on to the service.
Continue reading the main story
“Start Quote
Within moments, a rather scruffy figure loomed into view on my
phone - and I was video calling an old friend and colleague.”
End Quote Rory Cellan-Jones
* Skye video calls on test
Those participating machines act as what Skype calls "supernodes" and
carry out some of the administrative tasks of the global network and
help to ensure calls get through.
With a huge number of these machines offline because of the crash, the
rest of the network quickly became overloaded.
Mr Rabbe wrote that the disappearance of the supernodes meant the
remaining ones were swamped by traffic.
"The initial crashes happened just before our usual daily peak-hour and
very shortly after the initial crash," wrote Mr Rabbe, "which resulted
in traffic to the supernodes that was about 100 times what would
normally be expected at that time of day."
Traffic levels were so high that they blew through the safe operating
specifications supernodes usually use. As a result, more supernodes shut
down.
Compensation offer
The "confluence of events", said Mr Rabbe, led to Skype being offline
for about 24 hours as engineers put in place hundreds of dedicated
supernodes and gradually brought the service back to life.
To ensure the outage does not happen again, Mr Rabbe said Skype would
look at its update policy, to see if it should automatically move users
to newer versions of its software.
A version of Skype for Windows that is free of the bug already exists,
but is not automatically given to users.
It said it would also look at its network to improve capacity and get on
with an investment programme that would boost this resilience.
Mr Rabbe apologised again on behalf of the company and added: "We know
that we fell short in both fulfilling your expectations and
communicating with you during this incident."
Skype has offered compensation to customers in the form of vouchers for
pre-pay users and a free week of service for subscribers.
http://www.bbc.co.uk/news/technology-12092795
--
_/_/_/_/ _/_/_/_/_/ _/_/_/_/_/ Prof. Altamir Dias, Dr. Eng.
_/ _/ _/ _/ _/ Departamento de Eng. Mecânica
_/_/_/ _/ _/ _/ _/ UFSC Universidade Federal de SC
_/ _/ _/ _/ _/ 88.040-900 - Florianópolis-SC
_/_/_/_/ _/ _/ _/ _/_/_/_/_/ BRASIL
Phone: 55-48-3721-9264 - Ramal 210
Fax : 55-48-3721-7615
Voip : 55-48-3721-4001
http://www.emc.ufsc.br/professores/altamir/
Mais detalhes sobre a lista de discussão GUFSC