Ratul Mahajan

Bharathan Balaji, Brad Campbell, Amit Levy et al.

Internet of Things (IoT) promises to bring ease of monitoring, better efficiency and innovative services across many domains with connected devices around us. With information from critical parts of infrastructure and powerful cloud-based data analytics, many applications can be developed to gain insights about IoT systems as well as transform their capabilities. Actuation applications form an essential part of these IoT systems, as they enable automation as well as fast low-level decision making. However, modern IoT systems are designed for data acquisition, and actuation applications are implemented in an ad-hoc manner. We identify modeling constraints in a systematic manner as indispensable to support actuation applications because constraints encompass high-level policies dictated by laws of physics, legal policies, user preferences. We explore data models for constraints inIoT system with the example of a home heating system and illustrate the challenges in enforcing these constraints in theIoT system architecture.

Giacomo Bernardi, Ratul Mahajan, C. Seshadhri et al.

We design and deploy at Amazon the first production datacenter fabrics based on random graphs. While the cost and fault-tolerance benefits of such topologies have been long known, their practical realization has been hampered by a lack of scalable routing and cabling approaches. Our design, called RNG, has a new distributed routing protocol that exploits the properties of random graphs to find a large number of edge disjoint paths between endpoint pairs. A novel passive optical device that internally shuffles cable endpoints makes Amazon's cabling complexity similar to that of fat trees. We show that RNG fabrics match or exceed the performance of fat trees for a range of traffic patterns, despite being up to 45% cheaper. At Amazon, we made RNG the default datacenter fabric for most workloads.

Shubham Tiwari, Tapan Chugh, Nash Rickert et al.

Coding agents are a fast-growing LLM application, executing as long-running closed-loop sessions in which LLM generations alternate with external tool calls. Yet, unlike chat workloads, their serving behavior has not been studied extensively. We address this gap by collecting a dataset of real-world coding assistant traces. Our analysis shows that coding agent sessions repeatedly reuse large prefixes and create sustained KVCache pressure that conventional LLM serving policies handle poorly. Based on our analysis, we present CacheWise, a KVCache management layer that improves KVCache reuse for coding agent workloads. CacheWise combines prefix-aware scheduling with reuse-aware eviction guided by lightweight predictions from tool call metadata. Implemented in vLLM and evaluated on the collected traces, CacheWise reduces KVCache evictions by up to 2-2.6x and improves total agent session completion time by up to 3.5x.