Publications

This paper reviews an easily expandable plan for smart document handling across multiple cloud systems, aiming to make work easier to manage, more resilient to issues, and improve the total cost of ownership. The importance of this task stems from two factors: first, Intelligent Document Processing (IDP) tools are experiencing growth; second, multi-cloud use is expanding more widely. This increases the

Cloud service providers usually leverage standard benchmarks such as TPC-H and TPC-DS to evaluate and optimize the performance of cloud data analytic systems. However, these benchmarks have fixed query patterns and are unable to effectively generate statistics of the cloud workloads in production. For example, they cannot simulate the real workload with the similar performance metrics such as CPU Time and

NL2SQL (natural language to SQL) translates natural language questions into SQL queries, thereby making structured data accessible to non-technical users, serving as the foundation for intelligent data applications. State-of-the-art NL2SQL techniques typically perform translation by retrieving database-specific information, such as the database schema, and invoking a pre-trained large language model (LLM

We present insert-optimized implementations of three fundamental data sketching algorithms: Count Sketch (CS), SpaceSaving (SS), and Karnin-Lang-Liberty (KLL).While these sketches are widely used for approximate query processing and stream analytics, their practical insert performance often falls short of their full potential. Through careful engineering and novel implementation strategies, we achieve substantial

As the scale of training large language models (LLMs) increases, one emergent failure is silent data corruption (SDC), where hardware produces incorrect computations without explicit failure signals. In this work, we are the first to investigate the impact of real-world SDCs on LLM training by comparing model training between healthy production nodes and unhealthy nodes exhibiting SDCs. With the help from

In the classical problem formulation of Text-to-SQL in academia, each question is translated independently of the others into SQL. This differs from the setting in practice, where questions enter the Text-to-SQL system in sequence. Thus, for all but the first few questions, a translation history is available that contains past questions and how they were translated by the system. So far, it has not been

A metastable failure is a self-sustaining congestive collapse in which a system degrades in response to a transient stressor (e.g., a load surge) but fails to recover after the stressor is removed. These rare but potentially catastrophic events are notoriously hard to diagnose and mitigate, sometimes causing prolonged outages affecting millions of users. Ideally, we would discover susceptibility to metastable

Compute elasticity is a primary benefit of using cloud-based data processing platforms such as Amazon EMR, where clusters can be scaled both horizontally and vertically. For example, a query scanning petabytes of data can run faster in a cluster with thousands of nodes compared to one with only a few hundred. However, not all workloads require the same computational power or have the same resource utilization

We present the first technique to synthesize programs that compose side-effecting functions, pure functions, and control flow, from partial traces containing records of only the side-effecting functions. This technique can be applied to synthesize API composing scripts from logs of calls made to those APIs, or a script from traces of system calls made by a workload, for example. All of the provided traces

Hybrid models that combine the language modeling capabilities of Attention layers with the efficiency of Recurrent layers (e.g., State Space Models) have gained traction in practically supporting long contexts in Large Language Model serving. Yet, the unique properties of these models complicate the usage of complementary efficiency optimizations such as prefix caching that skip redundant computations across

Graph neural networks (GNNs) are models specialized for graph data and widely used in applications. To train GNNs on large graphs that exceed CPU memory, several systems have been designed to store data on disk and conduct out-of-core processing. However, these systems suffer from either read amplification when conducting random reads for node features that are smaller than a disk page, or degraded model

Recently the semiconductor industry has been alerted by hyperscaler companies reporting impact of field errors in megascale datacenters. They tend to be elusive and very difficult to detect until they affect a particular application several days or months after the IC has been deployed in a fleet. Although the cause of such errors can be manifold, ranging from test escapes and design marginalities to design

Large language models (LLMs) are ubiquitously powerful but prohibitively expensive to train, often requiring thousands of compute devices, typically GPUs. To reduce the cost of training LLMs for customers, Amazon Web Services (AWS) launched the Amazon EC2 trn1 instances, powered by AWS Trainium, Amazon’s homegrown deep-learning accelerator, as an alternative to distributed LLM training. The trn1 instances

There has been a growing demand for making modern cloud-based data analytics systems cost-effective and easy to use. AI-powered intelligent resource scaling is one such effort, aiming at automating scaling decisions for serverless offerings like Amazon Redshift Serverless. The foundation of intelligent resource scaling lies in the ability to forecast query workloads and their resource consumption accurately

Database performance troubleshooting is a complex multi-step process that broadly involves three key stages– (a) Detection: determining what’s wrong and when; (b) Root Cause Analysis (RCA): reasoning about why is the performance poor; (c) Resolution: identifying a fix. A plethora of techniques exist to address each of these problems, but they hardly work in real-world at scale. First, real-world customer