This paper explores the use of gossip-based protocols for computing aggregate information in highly distributed systems, which are characterized by high volatility and instability. The authors introduce the concept of *Node Aggregation* problems, where the goal is to compute aggregate functions (such as sums, averages, quantiles) from data held by multiple nodes in a network. They present the Push-Sum protocol, which is a simple and natural approach for computing sums or averages, and analyze its convergence properties. The paper also defines the notion of *diffusion speed*, which characterizes how quickly values diffuse through the network, and shows that this speed is crucial for the approximation guarantees of various protocols. The authors analyze the diffusion speed of Uniform Gossip and other communication mechanisms, including the impact of node failures and message loss. They also propose protocols for more complex queries, such as random sampling and quantile computation, and discuss practical considerations and future directions for research. The paper concludes with a discussion of the limitations of the approach and suggestions for improving convergence in certain network topologies.This paper explores the use of gossip-based protocols for computing aggregate information in highly distributed systems, which are characterized by high volatility and instability. The authors introduce the concept of *Node Aggregation* problems, where the goal is to compute aggregate functions (such as sums, averages, quantiles) from data held by multiple nodes in a network. They present the Push-Sum protocol, which is a simple and natural approach for computing sums or averages, and analyze its convergence properties. The paper also defines the notion of *diffusion speed*, which characterizes how quickly values diffuse through the network, and shows that this speed is crucial for the approximation guarantees of various protocols. The authors analyze the diffusion speed of Uniform Gossip and other communication mechanisms, including the impact of node failures and message loss. They also propose protocols for more complex queries, such as random sampling and quantile computation, and discuss practical considerations and future directions for research. The paper concludes with a discussion of the limitations of the approach and suggestions for improving convergence in certain network topologies.