Pinecone's Dedicated Read Nodes (DRN) are designed for high-demand applications that require consistent performance and low latency. They address the needs of workloads like billion-vector semantic search and high-QPS recommendation systems. With DRN, you get dedicated infrastructure that isolates queries from other workloads, eliminating issues like noisy neighbors and read rate limits. This ensures predictable performance and cost, which is crucial for applications with strict service-level objectives (SLOs). DRN scales through a combination of replicas and shards. Replicas enhance throughput and availability, while shards increase storage capacity. For example, each shard provides 250 GB of storage, and users can easily adjust the number of replicas and shards as their data needs change. Pinecone manages data movement automatically, so there's no complicated migration process required. The pricing model is hourly per node, making it easier to forecast costs, particularly for sustained high-query-per-second (QPS) workloads. Real-world performance highlights the effectiveness of DRN. One customer with 135 million vectors achieved 600 QPS with a P50 latency of 45ms and a P99 latency of 96ms. Another customer tested DRN with 1.4 billion vectors and reached an impressive 5,700 QPS at a P50 latency of 26ms. These examples illustrate how DRN handles demanding workloads while maintaining low latency and predictable performance. Setting up a DRN index involves selecting the appropriate node type, shards, and replicas in the Pinecone interface or via API. Users should ensure they understand the costs associated with provisioning dedicated hardware before proceeding. For those migrating from On-Demand to DRN, Pinecone provides API support to facilitate the transition.