[Paper Review] Optimal Gossip-Based Aggregate Computation

Jen‐Yeu Chen, Gopal Pandurangan|arXiv (Cornell University)|Jan 19, 2010

Peer-to-Peer Network Technologies19 references4 citations

TL;DR

This paper presents the first gossip-based protocol for aggregate computation that is both time-optimal and message-optimal, achieving $O(\log n)$ time and $O(n\log\log n)$ messages using a novel distributed random ranking (DRR) technique. The method partitions nodes into small trees, computes local aggregates at roots, and uses uniform gossip to converge on global results, breaking the $\Omega(n\log n)$ message barrier in the non-address-oblivious model.

ABSTRACT

We present the first provably almost-optimal gossip-based algorithms for aggregate computation that are both time optimal and message-optimal. Given a $n$-node network, our algorithms guarantee that all the nodes can compute the common aggregates (such as Min, Max, Count, Sum, Average, Rank etc.) of their values in optimal $O(\log n)$ time and using $O(n \log \log n)$ messages. Our result improves on the algorithm of Kempe et al. \cite{kempe} that is time-optimal, but uses $O(n \log n)$ messages as well as on the algorithm of Kashyap et al. \cite{efficient-gossip} that uses $O(n \log \log n)$ messages, but is not time-optimal (takes $O(\log n \log \log n)$ time). Furthermore, we show that our algorithms can be used to improve gossip-based aggregate computation in sparse communication networks, such as in peer-to-peer networks. The main technical ingredient of our algorithm is a technique called {\em distributed random ranking (DRR)} that can be useful in other applications as well. DRR gives an efficient distributed procedure to partition the network into a forest of (disjoint) trees of small size. Our algorithms are non-address oblivious. In contrast, we show a lower bound of $Ω(n\log n)$ on the message complexity of any address-oblivious algorithm for computing aggregates. This shows that non-address oblivious algorithms are needed to obtain significantly better message complexity. Our lower bound holds regardless of the number of rounds taken or the size of the messages used. Our lower bound is the first non-trivial lower bound for gossip-based aggregate computation and also gives the first formal proof that computing aggregates is strictly harder than rumor spreading in the address-oblivious model.

Motivation & Objective

• To design a gossip-based protocol for aggregate computation that achieves both time and message optimality in resource-constrained networks.
• To overcome the $\Omega(n\log n)$ message complexity lower bound that applies to address-oblivious algorithms.
• To develop a distributed randomized procedure (DRR) that partitions a network into small, disjoint trees for efficient local aggregation.
• To demonstrate that non-address-obliviousness is necessary to achieve $O(n\log\log n)$ message complexity.
• To improve performance in sparse networks like peer-to-peer and sensor networks through optimized message and time usage.

Proposed method

• Introduce a distributed random ranking (DRR) procedure to partition the network into a forest of small, disjoint trees of size $O(\log n)$.
• Perform local aggregate computation (e.g., sum, min, max) at the root of each tree using internal gossip within the tree.
• Initiate a uniform gossip protocol among the tree roots to propagate and converge on the global aggregate value.
• Use probabilistic analysis and Chernoff bounds to show that typical values spread efficiently across the network in each phase.
• Classify communication stages as dense or sparse and bound message propagation using union bounds and Markov’s inequality.
• Leverage negative correlation among random variables to derive high-probability bounds on the number of nodes learning each value.

Experimental results

Research questions

• RQ1 Can a gossip-based protocol achieve both $O(\log n)$ time and $O(n\log\log n)$ message complexity for aggregate computation?
• RQ2 Is it possible to break the $\Omega(n\log n)$ message complexity barrier in gossip-based aggregate computation, and if so, under what conditions?
• RQ3 What is the role of address-obliviousness in limiting message efficiency for gossip-based protocols?
• RQ4 Can a distributed random ranking (DRR) technique be used to partition a network into small trees to enable efficient local aggregation?
• RQ5 Does the proposed protocol outperform existing protocols in both time and message complexity in sparse or dynamic networks?

Key findings

• The proposed protocol achieves optimal $O(\log n)$ time complexity and $O(n\log\log n)$ message complexity for computing aggregates such as Min, Max, Sum, Count, and Average.
• The protocol uses a novel distributed random ranking (DRR) technique to partition the network into small trees, enabling efficient local aggregation before global convergence.
• The protocol is non-address oblivious, and the authors prove a $\Omega(n\log n)$ message complexity lower bound for any address-oblivious gossip-based protocol, showing that non-address-obliviousness is necessary for improved message efficiency.
• In dense communication phases, at most $6^{t+1}\log n$ nodes learn a typical value with high probability, satisfying the induction hypothesis for the analysis.
• In sparse phases, at least a constant fraction ($>1/6$) of typical values remain typical at the end, ensuring sustained progress toward full dissemination.
• The protocol improves upon prior work: it matches the time optimality of Kempe et al. [9] while reducing message complexity from $O(n\log n)$ to $O(n\log\log n)$, and improves upon Kashyap et al. [8] by achieving time optimality despite using $O(n\log\log n)$ messages.