[Paper Review] Network Coding for Computing Part I : Cut-Set Bounds

Rathinakumar Appuswamy, Massimo Franceschetti+2 more|arXiv (Cornell University)|Dec 15, 2009

Cooperative Communication and Network Coding7 citations

TL;DR

This paper introduces a network computing framework where a receiver computes a function of messages from source nodes, defining a computing capacity as the maximum rate of function computation per network use. It establishes a min-cut-based upper bound on this capacity and proves it is tight for linear functions and in multi-edge tree networks, while showing capacity can be arbitrarily small relative to the min-cut for certain functions and networks.

ABSTRACT

The following extit{network computing} problem is considered. Source nodes in a directed acyclic network generate independent messages and a single receiver node computes a target function $f$ of the messages. The objective is to maximize the average number of times $f$ can be computed per network usage, i.e., the ``computing capacity''. The extit{network coding} problem for a single-receiver network is a special case of the network computing problem in which all of the source messages must be reproduced at the receiver. For network coding with a single receiver, routing is known to achieve the capacity by achieving the network extit{min-cut} upper bound. We extend the definition of min-cut to the network computing problem and show that the min-cut is still an upper bound on the maximum achievable rate and is tight for computing (using coding) any target function in multi-edge tree networks and for computing linear target functions in any network. We also study the bound's tightness for different classes of target functions. In particular, we give a lower bound on the computing capacity in terms of the Steiner tree packing number and a different bound for symmetric functions. We also show that for certain networks and target functions, the computing capacity can be less than an arbitrarily small fraction of the min-cut bound.

Motivation & Objective

• To define and analyze the computing capacity in directed acyclic networks where a receiver computes a target function of source messages.
• To extend the concept of min-cut from traditional network coding to the network computing problem and establish it as an upper bound on computing capacity.
• To investigate the tightness of the min-cut bound across different classes of target functions, including linear and symmetric functions.
• To identify network and function conditions under which computing capacity can be significantly less than the min-cut bound.
• To provide lower bounds on computing capacity using Steiner tree packing and symmetry-based analysis.

Proposed method

• Defining the network computing problem as a generalization of network coding, where the receiver computes a function of source messages rather than reproducing them.
• Introducing a generalized min-cut definition for the network computing problem and proving it serves as an upper bound on the achievable computing rate.
• Analyzing the tightness of the min-cut bound in multi-edge tree networks, showing that coding achieves the bound for any target function.
• Proving that for linear target functions, the min-cut bound is tight in any network, extending the classic routing result to function computation.
• Deriving a lower bound on computing capacity using the Steiner tree packing number, which quantifies the maximum number of disjoint subnetworks supporting function computation.
• Introducing a symmetry-based bound for symmetric functions, showing improved capacity under symmetric structure in the network and function.

Experimental results

Research questions

• RQ1 Is the min-cut bound still valid as an upper limit on computing capacity in the network computing problem?
• RQ2 Under what network and function conditions is the min-cut bound tight?
• RQ3 Can computing capacity be arbitrarily smaller than the min-cut bound for certain functions and network topologies?
• RQ4 How do structural properties like symmetry or Steiner tree packing influence the lower bound on computing capacity?
• RQ5 To what extent do linear functions preserve the tightness of the min-cut bound in general networks?

Key findings

• The min-cut bound is a valid upper bound on the computing capacity for any target function in a directed acyclic network.
• The min-cut bound is tight for computing any target function in multi-edge tree networks, meaning coding achieves the theoretical maximum rate.
• For linear target functions, the min-cut bound is tight in any network, generalizing the classic routing result to function computation.
• There exist networks and target functions for which the computing capacity is less than an arbitrarily small fraction of the min-cut bound, demonstrating a significant gap.
• A lower bound on computing capacity is established in terms of the Steiner tree packing number, providing a structural measure of network capability.
• For symmetric functions, a separate lower bound is derived based on symmetry, which can improve capacity estimates in symmetric network-function settings.