arXiv subjects

Gourav Rattihalli

Publications and source records attributed to Gourav Rattihalli.

5 recordsLinked to original sources

arXiv2026

Predict, Reuse, and Repair: Accelerating Dynamic Sparse Attention for Long-Context LLM Decoding

Dynamic sparse attention (DSA) accelerates long-context LLM decoding by attending to only the top-K KV blocks relevant to each query, but it introduces a serialized selection-to-attention dependency that emerges as a new latency bottleneck. We present PRR, a speculate-reuse-repair runtime that exploits temporal locality in DSA selections to predict likely blocks, speculate the attention over them while selection is in flight, and incrementally repair missed blocks once the true selected set is known. PRR uses a lightweight EMA-based predictor, a profiling-guided speculation budget that keeps speculative work off the critical path, and a FlashAttention-based repair kernel that folds missed blocks into the partial attention state using online-softmax statistics. Across long-context benchmarks and representative DSA methods, PRR reduces per-token decoding latency by up to 40% while preserving downstream task accuracy. Github: https://github.com/Tianyu9748/Incremental_FlashAttention

cs.LG↗

arXiv2026

Energy-Aware Scheduling for Serverless LLM Serving on Shared GPUs

As LLM inference becomes a major cloud workload, its growing energy footprint makes cluster-wide energy optimization increasingly important. Serverless LLM serving helps platforms absorb traffic volatility by elastically sharing GPU resources across models, but this sharing also makes energy optimization difficult. Multiple co-resident models run under one device-wide operating point, while their resource demands and latency slack change across execution phases and load conditions. As a result, minimizing energy requires coordinated scheduling across request placement, runtime resource adaptation, and workload consolidation. We present Festina, a profiling-guided, power-aware control plane to minimize cluster-wide energy for serverless LLM serving. Unlike common global-local schedulers that focus on throughput or tail latency, Festina makes energy-first decisions by jointly coordinating request placement, SM partitioning, and GPU operating points under TTFT/TBT SLOs. In our system, a lightweight global scheduler performs fast, SLO-safe, energy-aware placement using constant-time lookups from offline profiles and GPU state summaries. On each GPU, a phase-aware local scheduler continuously adapts task batching and compute resources to minimize power consumption. Festina further performs energy-aware workload consolidation to reduce GPUs' static power consumption via SLO-aware migration. Comparison with four SOTA LLM serving systems and one DVFS-augmented system demonstrates that Festina reduces energy consumption by up to 56% while maintaining parity in SLO attainment (within a 2% margin)

cs.DC↗

arXiv2025

An RDMA-First Object Storage System with SmartNIC Offload

AI training and inference impose sustained, fine-grain I/O that stresses host-mediated, TCP-based storage paths. Motivated by kernel-bypass networking and user-space storage stacks, we revisit POSIX-compatible object storage for GPU-centric pipelines. We present ROS2, an RDMA-first object storage system design that offloads the DAOS client to an NVIDIA BlueField-3 SmartNIC while leaving the DAOS I/O engine unchanged on the storage server. ROS2 separates a lightweight control plane (gRPC for namespace and capability exchange) from a high-throughput data plane (UCX/libfabric over RDMA or TCP) and removes host mediation from the data path. Using FIO/DFS across local and remote configurations, we find that on server-grade CPUs RDMA consistently outperforms TCP for both large sequential and small random I/O. When the RDMA-driven DAOS client is offloaded to BlueField-3, end-to-end performance is comparable to the host, demonstrating that SmartNIC offload preserves RDMA efficiency while enabling DPU-resident features such as multi-tenant isolation and inline services (e.g., encryption/decryption) close to the NIC. In contrast, TCP on the SmartNIC lags host performance, underscoring the importance of RDMA for offloaded deployments. Overall, our results indicate that an RDMA-first, SmartNIC-offloaded object-storage stack is a practical foundation for scaling data delivery in modern LLM training environments; integrating optional GPU-direct placement for LLM tasks is left for future work.

cs.AR↗

arXiv2024

GreenFaaS: Maximizing Energy Efficiency of HPC Workloads with FaaS

Application energy efficiency can be improved by executing each application component on the compute element that consumes the least energy while also satisfying time constraints. In principle, the function as a service (FaaS) paradigm should simplify such optimizations by abstracting away compute location, but existing FaaS systems do not provide for user transparency over application energy consumption or task placement. Here we present GreenFaaS, a novel open source framework that bridges this gap between energy-efficient applications and FaaS platforms. GreenFaaS can be deployed by end users or providers across systems to monitor energy use, provide task-specific feedback, and schedule tasks in an energy-aware manner. We demonstrate that intelligent placement of tasks can both reduce energy consumption and improve performance. For a synthetic workload, GreenFaaS reduces the energy-delay product by 45% compared to alternatives. Furthermore, running a molecular design application through GreenFaaS can reduce energy consumption by 21% and runtime by 63% by better matching tasks with machines.

cs.DC↗

arXiv2019

Two stage cluster for resource optimization with Apache Mesos

As resource estimation for jobs is difficult, users often overestimate their requirements. Both commercial clouds and academic campus clusters suffer from low resource utilization and long wait times as the resource estimates for jobs, provided by users, is inaccurate. We present an approach to statistically estimate the actual resource requirement of a job in a Little cluster before the run in a Big cluster. The initial estimation on the little cluster gives us a view of how much actual resources a job requires. This initial estimate allows us to accurately allocate resources for the pending jobs in the queue and thereby improve throughput and resource utilization. In our experiments, we determined resource utilization estimates with an average accuracy of 90% for memory and 94% for CPU, while we make better utilization of memory by an average of 22% and CPU by 53%, compared to the default job submission methods on Apache Aurora and Apache Mesos.

cs.DC↗