Title: On the Capacity of Information Networks
1On the Capacity of Information Networks
- Nick Harvey
- Collaborators Micah Adler (UMass), Kamal Jain
(Microsoft), Bobby Kleinberg (MIT/Berkeley/Cornell
), and April Lehman (MIT/Google)
2What is the capacity of a network?
3What is the capacity of a network?
s2
s1
t1
t2
- Send items from s1?t1 and s2?t2
- Problem no disjoint paths
4An Information Network
s2
s1
t1
t2
- If sending information, we can do better
- Send xor b1?b2 on bottleneck edge
5Moral of Butterfly
Network Flow Capacity ? Information Flow Capacity
6Network Coding
- New approach for information flow problems
- Blend of combinatorial optimization, information
theory - Multicast, k-Pairs
- k-Pairs problems Network coding when each
commodity has one sink - Analogous to multicommodity flow
- Definitions for cyclic networks are subtle
7- Multicommodity
- Flow
- Efficient algorithms for computing maximum
concurrent (fractional) flow. - Connected with metric embeddings via LP duality.
- Approximate max-flow min-cut theorems.
- Network
- Coding
- Computing the max concurrent coding rate may be
- Undecidable
- Decidable in poly-time
- No adequate duality theory.
- No cut-based parameter is known to give sublinear
approximation in digraphs.
Directed and undirected problems behave quite
differently
8Directed k-pairs
- Coding rate can be muchlarger than flow rate!
- Butterfly
- Coding rate 1
- Flow rate ½
- Thm HKL04,LL04 ? graphs G(V,E) whereCoding
Rate O( flow rate V )
s1
s2
t2
t1
- Thm ? graphs G(V,E) whereCoding Rate O( flow
rate E ) - And this is optimal
- Recurse on butterfly construction
9Directed k-pairs
- Coding rate can be muchlarger than flow rate!
- and much larger than the sparsity(same
example)
Flow Rate ? Sparsity lt Coding Rate
in some graphs
10Undirected k-pairs
- No known undirected instance where coding rate ?
max flow rate! - (The undirected k-pairs conjecture)
Flow Rate ? Coding Rate ? Sparsity
Gap can be O(log n) when G is an expander
11Undirected k-Pairs Conjecture
Coding Rate
?
?
Sparsity
Flow Rate
Unknown until this work
Undirected k-pairs conjecture
12Okamura-Seymour Graph
Cut
Every cut has enough capacity to carry all
commodities separated by the cut
13Okamura-Seymour Max-Flow
Flow Rate 3/4
si is 2 hops from ti. At flow rate r, each
commodity consumes ? 2r units of bandwidth in a
graph with only 6 units of capacity.
14The trouble with information flow
s1
t3
- If an edge codes multiple commodities, how to
charge for consuming bandwidth? - We work around this obstacle and bound coding
rate by 3/4.
s2
s4
t1
t4
s3
t2
At flow rate r, each commodity consumes at least
2r units of bandwidth in a graph with only 6
units of capacity.
15Informational Dominance
- Definition A e if for every coding
solution,the messages sent on edges of A
uniquely determine the message sent on e. - Given A and e, how hard is it to determine
whether A e? Is it even decidable? - Theorem There is an algorithm tocompute whether
A e in time O(k²m). - Based on a combinatorial characterizationof
informational dominance
16What can we prove?
- Combine Informational Dominance with Shannon
inequalities for Entropy - Flow rate coding rate for Special Bipartite
Graphs - Bipartite
- Every source is 2 hopsaway from its sink
- Dual of flow LP is optimizedby assigning length
1 to all edges - Next show that proving conjecture for all graphs
is quite hard
s1 t3
s4
t4
s2 t1
s3 t2
17k-pairs conjecture I/O complexity
- I/O complexity model AV88
- A large, slow external memory consisting of pages
each containing p records - A fast internal memory that holds 2 pages
- Basic I/O operation read in two pages from
external memory, write out one page
18I/O Complexity of Matrix Transposition
- Matrix transposition Given a pp matrix of
records in row-major order, write it out in
column-major order. - Obvious algorithm requires O(p²) ops.
- A better algorithm uses O(p log p) ops.
19I/O Complexity of Matrix Transposition
- Matrix transposition Given a pp matrix of
records in row-major order, write it out in
column-major order. - Obvious algorithm requires O(p²) ops.
- A better algorithm uses O(p log p) ops.
s1
s2
20I/O Complexity of Matrix Transposition
- Matrix transposition Given a pxp matrix of
records in row-major order, write it out in
column-major order. - Obvious algorithm requires O(p²) ops.
- A better algorithm uses O(p log p) ops.
s1
s2
s3
s4
21I/O Complexity of Matrix Transposition
- Matrix transposition Given a pxp matrix of
records in row-major order, write it out in
column-major order. - Obvious algorithm requires O(p²) ops.
- A better algorithm uses O(p log p) ops.
s1
s2
s3
s4
t3
t1
22I/O Complexity of Matrix Transposition
- Matrix transposition Given a pxp matrix of
records in row-major order, write it out in
column-major order. - Obvious algorithm requires O(p²) ops.
- A better algorithm uses O(p log p) ops.
s1
s2
s3
s4
t3
t4
t1
t2
23Matching Lower Bound
- Theorem (Floyd 72, AV88) A matrix
transposition algorithm using only read and write
operations (no arithmetic on values) must perform
O(p log p) I/O operations.
s1
s2
s3
s4
t3
t4
t1
t2
24O(p log p) Lower Bound
- Proof Let Nij denote the number of ops in which
record (i,j) is written. For all j, - Si Nij p log p.
- Hence
- Sij Nij p² log p.
- Each I/O writes only p records. QED.
s1
s2
s3
s4
t3
t4
t1
t2
25The k-pairs conjecture and I/O complexity
- Definition An oblivious algorithm is one whose
pattern of read/write operations does not depend
on the input. - Theorem If there is an oblivious algorithm for
matrix transposition using o(p log p) I/O ops,
the undirected k-pairs conjecture is false.
s1
s2
s3
s4
t3
t4
t1
t2
26The k-pairs conjecture and I/O complexity
- Proof
- Represent the algorithm with a diagram as before.
- Assume WLOG that each node has only two outgoing
edges.
s1
s2
s3
s4
t3
t4
t1
t2
27The k-pairs conjecture and I/O complexity
- Proof
- Represent the algorithm with a diagram as before.
- Assume WLOG that each node has only two outgoing
edges. - Make all edges undirected, capacity p.
- Create a commodity for each matrix entry.
s1
s2
s3
s4
t3
t4
t1
t2
28The k-pairs conjecture and I/O complexity
- Proof
- The algorithm itself is a network code of rate 1.
- Assuming the k-pairs conjecture, there is a flow
of rate 1. - Si,jd(si,tj) p E(G).
- Arguing as before, LHS is O(p² log p).
- Hence E(G)O(p log p).
s1
s2
s3
s4
t3
t4
t1
t2
29Other consequences for complexity
- The undirected k-pairs conjecture implies
- A O(p log p) lower bound for matrix transposition
in the cell-probe model. - Same proof.
- A O(p² log p) lower bound for the running time of
oblivious matrix transposition algorithms on a
multi-tape Turing machine. - I/O model can emulate multi-tape Turing
machines with a factor p speedup.
30Distance arguments
- Rate-1 flow solution implies Si d(si,ti) E
- LP duality directed or undirected
- Does rate-1 coding solution implySi d(si,ti)
E? - Undirected graphs this is essentially
thek-pairs conjecture! - Directed graphs this is completely false
31Recursive construction
- k commodities (si,ti)
- Distance d(si,ti) O(log k) ?i
- O(k) edges!
32Recursive Construction
s1
s2
2 commodities 7 edges Distance 3
G (1)
t1
t2
s2
s1
Edge capacity 1
t1
t2
33Recursive Construction
s3
s4
s1
s2
G (2)
t3
t4
t1
t2
- Start with two copies of G (1)
34Recursive Construction
s3
s4
s1
s2
G (2)
t3
t4
t1
t2
- Replace middle edges with copy of G (1)
35Recursive Construction
s3
s4
s1
s2
G (1)
G (2)
t3
t4
t1
t2
- 4 commodities, 19 edges, Distance 5
36Recursive Construction
s1
s2
s3
s4
s2n-1
s2n
G (n-1)
G (n)
t1
t2
t3
t4
t2n-1
t2n
- commodities 2n, V O(2n), E O(2n)
- Distance 2n1
37Summary
- Directed instances
- Coding rate gtgt flow rate
- Undirected instances
- Conjecture Flow rate Coding rate
- Proof for special bip graphs
- Tool Informational Dominance
- Proving conjecture solves MatrixTransposition
Problem
38Open Problems
- Computing the network coding rate in DAGs
- Recursively decidable?
- How do you compute a o(n)-factor approximation?
- Undirected k-pairs conjecture
- Stronger complexity consequences?
- Prove a O(log n) gap between sparsest cut and
coding rate for some graphs - or, find a fast matrix transposition algorithm.
39Backup Slides
40Optimality
- The graph G (n) provesThm HKL05 ? graphs
G(V,E) whereNCR O( flow rate E ) - G (n) is optimalThm HKL05 ? graph
G(V,E),NCR/flow rate O(min V,E,k)
41Informational Dominance
- Def A dominates B if information in A determines
information in Bin every network coding solution.
s1
s2
A does not dominate B
t2
t1
42Informational Dominance
- Def A dominates B if information in A determines
information in Bin every network coding solution.
s1
s2
A dominates B
Sufficient Condition If no path from any source
? B then A dominates B
t2
t1
43Informational Dominance Example
- Obviously flow rate NCR 1
- How to prove it? Markovicity?
- No two edges disconnect t1 and t2 from both
sources!
44Informational Dominance Example
s1
s2
t1
Cut A
t2
Sufficient Condition If no path from any source
? B then A dominates B
45Informational Dominance Example
s1
s2
t1
Cut A
t2
- Our characterization implies thatA dominates
t1,t2 ? H(A) ? H(t1,t2)
46Rate ¾ for Okamura-Seymour
s1
s1
t3
s1 t3
i
s4
t4
s2 t1
s3
s3 t2
47Rate ¾ for Okamura-Seymour
s1 t3
i
s4
t4
s2 t1
i
s3 t2
i
48Rate ¾ for Okamura-Seymour
s1 t3
i
s4
t4
s2 t1
i
s3 t2
i
49Rate ¾ for Okamura-Seymour
s1 t3
s4
t4
s2 t1
s3 t2
50Rate ¾ for Okamura-Seymour
s1 t3
s4
t4
s2 t1
s3 t2
51Rate ¾ for Okamura-Seymour
s1 t3
s4
t4
s2 t1
¾ RATE
3 H(source) 6 H(undirected edge) 11 H(source)
6 H(undirected edge) 8 H(source)
s3 t2