arXiv subjects

Akshay Jajoo

Publications and source records attributed to Akshay Jajoo.

10 recordsLinked to original sources

arXiv2026

KLAS: Using Similarity to Stitch Neural Networks for Improved Accuracy-Efficiency Tradeoffs

Given the wide range of deployment targets, flexible model selection is essential for optimizing performance within a given compute budget. Recent work demonstrates that stitching pretrained models within a model family enables cost-effective interpolation of the accuracy-efficiency tradeoff space. Stitching transforms intermediate activations from one pretrained model into another, producing a new interpolated stitched network. Such networks provide a pool of deployment options along the accuracy-efficiency spectrum. However, existing stitching approaches often yield suboptimal tradeoffs and lack generalizability, as they primarily rely on heuristics to select stitch configurations. We argue that constructing improved accuracy-efficiency tradeoffs requires explicitly capturing and leveraging the similarity between pretrained models being stitched. To this end, we introduce KLAS, a novel stitch selection framework that automates and generalizes stitch selection across model families by leveraging KL divergence between intermediate representations. KLAS identifies the most promising binary stitches from the $O(k^2n^2)$ possibilities for $k$ pretrained models of depth $n$. Through comprehensive experiments, we demonstrate that KLAS improves the accuracy-efficiency curve of stitched models at the same finetuning cost as baselines. KLAS achieves up to $1.21\%$ higher ImageNet-1K top-1 accuracy at the same computational cost, or maintains accuracy with a $1.33\times$ reduction in FLOPs.

cs.LG↗

arXiv2026

Cornfigurator: Automated Planning for Any-to-Any Multimodal Model Serving

Any-to-Any models are an emerging class of multimodal models that accept combinations of text and multimodal data as input and generate them as output, introducing heterogeneous computation paths and component scaling characteristics. There are existing mechanisms for deploying Any-to-Any models--or special cases of them--for inference serving, but they either require manual effort and expertise to tune, or do not generalize to generic Any-to-Any models. We present Cornfigurator, the first deployment planner for generic Any-to-Any model inference serving. The goal of Cornfigurator is to maximize the overall goodput of serving the model, defined as the throughput of requests meeting their latency targets. To do so, based on model and workload characteristics, Cornfigurator explores the full spectrum of deployment strategies, from colocation to disaggregation and mixing different strategies. Cornfigurator performs coarse-to-fine statistical evaluation to efficiently navigate the large space of candidate plans. Plans generated by Cornfigurator either match or deliver 1.12$\times$-6.32$\times$ higher goodput compared to existing systems and expert-tuned deployment plans.

cs.LG↗

arXiv2026

Cornserve: A Distributed Serving System for Any-to-Any Multimodal Models

Any-to-Any models are an emerging class of multimodal models that accept combinations of multimodal data (e.g., text, image, video, audio) as input and generate them as output. Serving these models are challenging; different requests with different input and output modalities traverse different paths through the model computation graph, and each component of the model have different scaling characteristics. We present Cornserve, a distributed serving system for generic Any-to-Any models. Cornserve provides a flexible task abstraction for expressing Any-to-Any model computation graphs, enabling component disaggregation and independent scaling. The distributed runtime dispatches compute to the data plane via an efficient record-and-replay execution model that keeps track of data dependencies, and forwards tensor data between components directly from the producer to the consumer. Built on Kubernetes with approximately 23K new lines of Python, Cornserve supports diverse Any-to-Any models and delivers up to 3.81$\times$ higher throughput and 5.79$\times$ lower tail latency. Cornserve is open-source, and the demo video is available on YouTube.

cs.LG↗

arXiv2024

BlockLLM: Multi-tenant Finer-grained Serving for Large Language Models

The increasing demand for Large Language Models (LLMs) across various applications has led to a significant shift in the design of deep learning serving systems. Deploying LLMs, particularly in multi-tenant environments, poses substantial challenges due to their high computational and memory demands. We introduce BlockLLM, a serving system that leverages component sharing among fine-tuned LLM models to provide an efficient and flexible solution for LLM workloads. BlockLLM partitions models into finer-grained blocks, enabling the reuse of model components and independent provisioning to improve computation efficiency. BlockLLM comprises an offline block zoo for storing blocks and an online system to serve requests through chains of blocks. It offers multi-fold flexibilities: (1) Adaptive assembly of blocks on-the-fly through equivalence evaluation among blocks in the zoo; (2) Per-block batch size configuration and best-effort KV cache coordination at the individual block level; (3) Speculative execution and locality-aware block placement to reduce communication costs from dynamic block resource allocation. Our evaluation shows that BlockLLM reduces memory and storage footprints and improves computational efficiency, outperforming existing serving approach in 95%ile latency and GPU utilization by 33.5% and 20.1%, respectively, with minimal impact on accuracy

cs.DC↗

arXiv2024

Enabling Elastic Model Serving with MultiWorld

Machine learning models have been exponentially growing in terms of their parameter size over the past few years. We are now seeing the rise of trillion-parameter models. The large models cannot fit into a single GPU and thus require partitioned deployment across GPUs and even hosts. A high-performance collective communication library (CCL) such as NCCL is essential to fully utilize expensive GPU resources. However, CCL is not a great fit for inference. Unlike training for which a fixed amount of GPU resources is used for fixed workloads (e.g., input datasets), the inference workloads can change dynamically over time. Failures at the serving time can also impact individual user's experiences directly. In contrast, workers in a CCL process group share a single fault domain and the process group cannot grow as the workloads increase. The gap between the unique characteristics of model serving and CCL's nature makes it hard to serve large models elastically. To bridge the gap, we propose MultiWorld that enables fault tolerance and online scaling at the granularity of workers for model serving. Our evaluation showcases that enabling these new functionalities incurs small overheads (1.4-4.3% throughput loss) for most of the scenarios we tested.

cs.DC↗

arXiv2024

Atomicity and Abstraction for Cross-Blockchain Interactions

A blockchain facilitates secure and atomic transactions between mutually untrusting parties on that chain. Today, there are multiple blockchains with differing interfaces and security properties. Programming in this multi-blockchain world is hindered by the lack of general and convenient abstractions for cross-chain communication and computation. Current cross-chain communication bridges have varied and low-level interfaces, making it difficult to develop portable applications. Current methods for multi-chain atomic transactions are limited in scope to cryptocurrency swaps. This work addresses these issues. We first define a uniform, high-level interface for communication between chains. Building on this interface, we formulate a protocol that guarantees atomicity for general transactions whose operations may span several chains. We formulate and prove the desired correctness and security properties of these protocols. Our prototype implementation is built using the LayerZero cross-chain bridge. Experience with this implementation shows that the new abstractions considerably simplify the design and implementation of multi-chain transactions. Experimental evaluation with multi-chain swap transactions demonstrates performance comparable to that of custom-built implementations.

cs.CR↗

arXiv2022

A Case for Sampling Based Learning Techniques in Coflow Scheduling

Coflow scheduling improves data-intensive application performance by improving their networking performance. State-of-the-art online coflow schedulers in essence approximate the classic Shortest-Job-First (SJF) scheduling by learning the coflow size online. In particular, they use multiple priority queues to simultaneously accomplish two goals: to sieve long coflows from short coflows, and to schedule short coflows with high priorities. Such a mechanism pays high overhead in learning the coflow size: moving a large coflow across the queues delays small and other large coflows, and moving similar-sized coflows across the queues results in inadvertent round-robin scheduling. We propose Philae, a new online coflow scheduler that exploits the spatial dimension of coflows, i.e., a coflow has many flows, to drastically reduce the overhead of coflow size learning. Philae pre-schedules sampled flows of each coflow and uses their sizes to estimate the average flow size of the coflow. It then resorts to Shortest Coflow First, where the notion of shortest is determined using the learned coflow sizes and coflow contention. We show that the sampling-based learning is robust to flow size skew and has the added benefit of much improved scalability from reduced coordinator-local agent interactions. Our evaluation using an Azure testbed, a publicly available production cluster trace from Facebook shows that compared to the prior art Aalo, Philae reduces the coflow completion time (CCT) in average (P90) cases by 1.50x (8.00x) on a 150-node testbed and 2.72x (9.78x) on a 900-node testbed. Evaluation using additional traces further demonstrates Philae's robustness to flow size skew.

cs.DC↗

arXiv2021

A study on the Morris Worm

The Morris worm was one of the first worms spread via the internet. It was spread on November 2, 1988, and changed how computer security was viewed by computer professionals as well as general public. Since its inception the Morris worm has been studied extensively from the security point of view and is still a point of interest. This paper summarizes the effects, impacts, and lessons learned from the episode. There are other copies of this paper present. However, I recommend using the arXiv version only.

cs.CR↗

arXiv2021

The Case for Task Sampling based Learning for Cluster Job Scheduling

The ability to accurately estimate job runtime properties allows a scheduler to effectively schedule jobs. State-of-the-art online cluster job schedulers use history-based learning, which uses past job execution information to estimate the runtime properties of newly arrived jobs. However, with fast-paced development in cluster technology (in both hardware and software) and changing user inputs, job runtime properties can change over time, which lead to inaccurate predictions. In this paper, we explore the potential and limitation of real-time learning of job runtime properties, by proactively sampling and scheduling a small fraction of the tasks of each job. Such a task-sampling-based approach exploits the similarity among runtime properties of the tasks of the same job and is inherently immune to changing job behavior. Our study focuses on two key questions in comparing task-sampling-based learning (learning in space) and history-based learning (learning in time): (1) Can learning in space be more accurate than learning in time? (2) If so, can delaying scheduling the remaining tasks of a job till the completion of sampled tasks be more than compensated by the improved accuracy and result in improved job performance? Our analytical and experimental analysis of 3 production traces with different skew and job distribution shows that learning in space can be substantially more accurate. Our simulation and testbed evaluation on Azure of the two learning approaches anchored in a generic job scheduler using 3 production cluster job traces shows that despite its online overhead, learning in space reduces the average Job Completion Time (JCT) by 1.28x, 1.56x, and 1.32x compared to the prior-art history-based predictor.

cs.DC↗

arXiv2021

Saath: Speeding up CoFlows by Exploiting the Spatial Dimension

Coflow scheduling improves data-intensive application performance by improving their networking performance. State-of-the-art Coflow schedulers in essence approximate the classic online Shortest-Job-First (SJF) scheduling, designed for a single CPU, in a distributed setting, with no coordination among how the flows of a Coflow at individual ports are scheduled, and as a result suffer two performance drawbacks: (1) The flows of a Coflow may suffer the out-of-sync problem -- they may be scheduled at different times and become drifting apart, negatively affecting the Coflow completion time (CCT); (2) FIFO scheduling of flows at each port bears no notion of SJF, leading to suboptimal CCT. We propose SAATH, an online Coflow scheduler that overcomes the above drawbacks by explicitly exploiting the spatial dimension of Coflows. In SAATH, the global scheduler schedules the flows of a Coflow using an all-or-none policy which mitigates the out-of-sync problem. To order the Coflows within each queue, SAATH resorts to a Least-Contention-First (LCoF) policy which we show extends the gist of SJF to the spatial dimension, complemented with starvation freedom. Our evaluation using an Azure testbed and simulations of two production cluster traces show that compared to Aalo, SAATH reduces the CCT in median (P90) cases by 1.53x (4.5x) and 1.42x (37x), respectively.

cs.DC↗