arXiv subjects

Shashwat Jaiswal

Publications and source records attributed to Shashwat Jaiswal.

3 recordsLinked to original sources

arXiv2025

Serving Heterogeneous LoRA Adapters in Distributed LLM Inference Systems

Low-Rank Adaptation (LoRA) has become the de facto method for parameter-efficient fine-tuning of large language models (LLMs), enabling rapid adaptation to diverse domains. In production, LoRA-based models are served at scale, creating multi-tenant environments with hundreds of adapters sharing a base model. However, state-of-the-art serving systems co-batch heterogeneous adapters without accounting for rank (size) variability, leading to severe performance skew, which ultimately requires adding more GPUs to satisfy service-level objectives (SLOs). Existing optimizations, focused on loading, caching, and kernel execution, ignore this heterogeneity, leaving GPU resources underutilized. We present LoRAServe, a workload-aware dynamic adapter placement and routing framework designed to tame rank diversity in LoRA serving. By dynamically rebalancing adapters across GPUs and leveraging GPU Direct RDMA for remote access, LoRAServe maximizes throughput and minimizes tail latency under real-world workload drift. Evaluations on production traces from Company X show that LoRAServe elicits up to 2$\times$ higher throughput, up to 9$\times$ lower TTFT, while using up to 50% fewer GPUs under SLO constraints compared to state-of-the-art systems.

cs.DC↗

arXiv2025

SageServe: Optimizing LLM Serving on Cloud Data Centers with Forecast Aware Auto-Scaling

Global cloud service providers handle inference workloads for Large Language Models (LLMs) that span latency-sensitive (e.g., chatbots) and insensitive (e.g., report writing) tasks, resulting in diverse and often conflicting Service Level Agreement (SLA) requirements. Managing such mixed workloads is challenging due to the complexity of the inference serving stack, which encompasses multiple models, GPU hardware, and global data centers. Existing solutions often silo such fast and slow tasks onto separate GPU resource pools with different SLAs, but this leads to significant under-utilization of expensive accelerators due to load mismatch. In this article, we characterize the LLM serving workloads at Microsoft Office 365, one of the largest users of LLMs within Microsoft Azure cloud with over 10 million requests per day, and highlight key observations across workloads in different data center regions and across time. This is one of the first such public studies of Internet-scale LLM workloads. We use these insights to propose SageServe, a comprehensive LLM serving framework that dynamically adapts to workload demands using multi-timescale control knobs. It combines short-term request routing to data centers with long-term scaling of GPU VMs and model placement with higher lead times, and co-optimizes the routing and resource allocation problem using a traffic forecast model and an Integer Linear Programming (ILP) solution. We evaluate SageServe through real runs and realistic simulations on 10 million production requests across three regions and four open-source models. We achieve up to 25% savings in GPU-hours compared to the current baseline deployment and reduce GPU-hour wastage due to inefficient auto-scaling by 80%, resulting in a potential monthly cost savings of up to $2.5 million, while maintaining tail latency and meeting SLAs.

cs.DC↗

arXiv2025

AerialDB: A Federated Peer-to-Peer Spatio-temporal Edge Datastore for Drone Fleets

Recent years have seen an unprecedented growth in research that leverages the newest computing paradigm of Internet of Drones, comprising a fleet of connected Unmanned Aerial Vehicles (UAVs) used for a wide range of tasks such as monitoring and analytics in highly mobile and changing environments characteristic of disaster regions. Given that the typical data (i.e., videos and images) collected by the fleet of UAVs deployed in such scenarios can be considerably larger than what the onboard computers can process, the UAVs need to offload their data in real-time to the edge and the cloud for further processing. To that end, we present the design of AerialDB - a lightweight decentralized data storage and query system that can store and process time series data on a multi-UAV system comprising: A) a fleet of hundreds of UAVs fitted with onboard computers, and B) ground-based edge servers connected through a cellular link. Leveraging lightweight techniques for content-based replica placement and indexing of shards, AerialDB has been optimized for efficient processing of different possible combinations of typical spatial and temporal queries performed by real-world disaster management applications. Using containerized deployment spanning up to 400 drones and 80 edges, we demonstrate that AerialDB is able to scale efficiently while providing near real-time performance with different realistic workloads. Further, AerialDB comprises a decentralized and locality-aware distributed execution engine which provides graceful degradation of performance upon edge failures with relatively low latency while processing large spatio-temporal data. AerialDB exhibits comparable insertion performance and 100 times improvement in query performance against state-of-the-art baseline. Moreover, it exhibits a 10 times and 100 times improvement with insertion and query workloads respectively over the cloud baseline.

cs.DC↗