Rethinking Identity Management for
Large-scale Distributed Systems
Weverton Luis da Costa Cordeiro weverton cordeiro@inf ufrgs br [email protected]
Agenda
• Identity management todayy g y • What we think about it…
• What we see (as prominent solutions) about it… • What we envisage about it…
• Our framework for it… • The road ahead for it…
Identity management today…
Online systems offer a lightweight processy g g p so that user can create identities
Such convenience comes with a price: with minimum Such convenience comes with a price: with minimum
Identity management today…
Fake accounts can serve for a variety of purposes, y p p which mainly depend on the system nature
Promote people and advertisements
Promote words and tweets
Identity management today…
• The problem of creation of fake accounts is toughp g
• No solution can be achieved without some degree of centralization (Douceur, 2002)
centralization (Douceur, 2002)
• There can be no “one-size-fits-all” solution
i t ith di it f i t
• environments with a diversity of requirements
• varied user constraints
• Existing solutions fall in either of two categories
• Strong-based identity schemes
What we think about identity
management
management…
• It should preserve the users’ privacyp p y
• Obtaining a single identity
• should be cheap for honest users
What we see as prominent directions
for identity management
for identity management…
CAPTCHAs
The attacker can redirect them for some other user to solve for some other user to solve
Computational puzzles
Existing approaches use puzzles of fixed capacity
Waste of processing cycles for computing useless information
What we envisage as a prominent
framework for identity management
framework for identity management…
• Focus on those distributed systemsy based on the peer-to-peer paradigm
W k id tit h ( t th ti ti )
• Weak identity schemes (no strong authentication)
We hypothesize that the attacker recurs much more • We hypothesize that the attacker recurs much more
frequently than a honest user to obtain identities
• Tracking back these identity requests to their
What we envisage as a prominent
framework for identity management
framework for identity management…
Identity requests Bootstrap y Bootstrap service Identity requests Bootstrap service service
What we envisage as a prominent
framework for identity management
framework for identity management…
Identity requests Bootstrap y Bootstrap service Identity requests Bootstrap service service
The challenge is that sources cannot be reliably tracked, and may change overtime
What we envisage as a prominent
framework for identity management
What we envisage as a prominent
framework for identity management
framework for identity management…
Trust score Trust score
Trust score Trust score
Trust score
Trust score Trust score
Trust score
What we envisage as a prominent
framework for identity management
framework for identity management…
Trust score Trust score
Trust score Trust score
Trust score
Trust score Trust score
Trust score
What we envisage as a prominent
framework for identity management
framework for identity management…
Trust score Trust score
Trust score Trust score
Trust score
Trust score Trust score
Trust score
Our framework…
• Preserves the user privacy (primary goal)p y (p y g )
• Evaluates the likeliness of an ongoing attack during
th d i i h ( b h i l i )
the admission phase (no user behavior analysis)
Relies on the concept of sources of identity requests • Relies on the concept of sources of identity requests,
not on the strategies that can materialize them
• Can accommodate any strategy for pricing identity requests (in terms of effort or computational cost)
Our framework…
Trust Scores (CNSM 2011)
( )
• Trust score derived from a relationship between the source and
Our framework…
Trust Scores (CNSM 2011) Adaptive Puzzles (COMNET 2012) ( ) ( )• Trust score derived from a relationship between the source and
network recurrence rates
• Puzzles of adaptive complexity defined as a function of the
Our framework…
Trust Scores (CNSM 2011) Adaptive Puzzles (COMNET 2012) Green and Useful Puzzles ( ) ( ) (IM 2013)• Trust score derived from a relationship between the source and
network recurrence rates
• Puzzles of adaptive complexity defined as a function of the
source trust score, and assigned to users
• Traditional puzzles reshaped into simulation jobs (following theTraditional puzzles reshaped into simulation jobs (following the
ReCAPTCHA project philosophy) and wait time
• reduces energy consumption
• make puzzle-processing useful
Our framework…
Trust Scores (CNSM 2011) Adaptive Puzzles (COMNET 2012) Green and Useful Puzzles Future (?) ( ) ( ) (IM 2013)• Trust score derived from a relationship between the source and
network recurrence rates
( )
• Puzzles of adaptive complexity defined as a function of the
source trust score, and assigned to users
• Traditional puzzles reshaped into simulation jobs (following theTraditional puzzles reshaped into simulation jobs (following the
ReCAPTCHA project philosophy) and wait time
• reduces energy consumption
• make puzzle-processing useful
The road ahead…
Existing challenges at this stage
Existing challenges at this stage
1 design a decentralized approach e g for 1. design a decentralized approach, e.g. for
super-peers and chord-based architectures
2. obtain recent network profiles to use as 2. obtain recent network profiles to use as
input data for simulation and analysis
3. evaluate our framework in-the-wild
(modeling, simulation, and planetlab so far)
4. coexist with existing identity management
f k ( th b d t ti )
Rethinking Identity Management for
Large-scale Distributed Systems
Weverton Luis da Costa Cordeiro weverton cordeiro@inf ufrgs br [email protected]
Eclipse Attack…
Fake accounts can serve for a variety of purposes, y p p which mainly depend on the system nature
Eclipse Attack…
Fake accounts can serve for a variety of purposes, y p p which mainly depend on the system nature
Eclipse Attack…
Fake accounts can serve for a variety of purposes, y p p which mainly depend on the system nature
Eclipse Attack…
Fake accounts can serve for a variety of purposes, y p p which mainly depend on the system nature
Analytical Model
• We demonstrated some important properties that our p p p framework holds (to appear)
• Given a set of resources in hands of an attacker, she can only maximize the profit of her attack by evenly dividing the
maximize the profit of her attack by evenly dividing the
number of identities to be requested among these resources
• Given a time period for the attack, the attacker can only
maximize the profit of her attack by uniformly distributing the maximize the profit of her attack by uniformly distributing the identity requests to be performed throughout that period
• An attacker, sharing sources with honest users, causes minimal impact to their requests; instead, such source sharing even degrades further the attack profit
Results with simulation
160,000 honest users trying to obtain identities over one week
An attacker, using varying proportions of sources trying proportions of sources, trying to get 80,000 identities in the same period
Results with PlanetLab
240 honest users trying to obtain 2,400 identities over two hours
An attacker, using 20 sources, trying to get 1 200 identities in trying to get 1,200 identities in the same period