arXiv subjects

Ana Klimovic

Publications and source records attributed to Ana Klimovic.

30 records · Page 2Linked to original sources

arXiv2024

Boxer: FaaSt Ephemeral Elasticity for Off-the-Shelf Cloud Applications

Elasticity is a key property of cloud computing. However, elasticity is offered today at the granularity of virtual machines, which take tens of seconds to start. This is insufficient to react to load spikes and sudden failures in latency sensitive applications, leading users to resort to expensive overprovisioning. Function-as-a-Service (FaaS) provides significantly higher elasticity than VMs, but comes coupled with an event-triggered programming model and a constrained execution environment that makes them unsuitable for off-the-shelf applications. Previous work tries to overcome these obstacles but often requires re-architecting the applications. In this paper, we show how off-the-shelf applications can transparently benefit from ephemeral elasticity with FaaS. We built Boxer, an interposition layer spanning VMs and AWS Lambda, that intercepts application execution and emulates the network-of-hosts environment that applications expect when deployed in a conventional VM/container environment. The ephemeral elasticity of Boxer enables significant performance and cost savings for off-the-shelf applications with, e.g., recovery times over 5x faster than EC2 instances and absorbing load spikes comparable to overprovisioned EC2 VM instances.

cs.DC↗

arXiv2024

Imaginary Machines: A Serverless Model for Cloud Applications

Serverless Function-as-a-Service (FaaS) platforms provide applications with resources that are highly elastic, quick to instantiate, accounted at fine granularity, and without the need for explicit runtime resource orchestration. This combination of the core properties underpins the success and popularity of the serverless FaaS paradigm. However, these benefits are not available to most cloud applications because they are designed for networked virtual machines/containers environments. Since such cloud applications cannot take advantage of the highly elastic resources of serverless and require run-time orchestration systems to operate, they suffer from lower resource utilization, additional management complexity, and costs relative to their FaaS serverless counterparts. We propose Imaginary Machines, a new serverless model for cloud applications. This model (1.) exposes the highly elastic resources of serverless platforms as the traditional network-of-hosts model that cloud applications expect, and (2.) it eliminates the need for explicit run-time orchestration by transparently managing application resources based on signals generated during cloud application executions. With the Imaginary Machines model, unmodified cloud applications become serverless applications. While still based on the network-of-host model, they benefit from the highly elastic resources and do not require runtime orchestration, just like their specialized serverless FaaS counterparts, promising increased resource utilization while reducing management costs.

cs.DC↗

arXiv2024

DéjàVu: KV-cache Streaming for Fast, Fault-tolerant Generative LLM Serving

Distributed LLM serving is costly and often underutilizes hardware accelerators due to three key challenges: bubbles in pipeline-parallel deployments caused by the bimodal latency of prompt and token processing, GPU memory overprovisioning, and long recovery times in case of failures. In this paper, we propose DéjàVu, a system to address all these challenges using a versatile and efficient KV cache streaming library (DéjàVuLib). Using DéjàVuLib, we propose and implement efficient prompt-token disaggregation to reduce pipeline bubbles, microbatch swapping for efficient GPU memory management, and state replication for fault-tolerance. We highlight the efficacy of these solutions on a range of large models across cloud deployments.

cs.DC↗

arXiv2024

tf.data service: A Case for Disaggregating ML Input Data Processing

Machine learning (ML) computations commonly execute on expensive specialized hardware, such as GPUs and TPUs, which provide high FLOPs and performance-per-watt. For cost efficiency, it is essential to keep these accelerators highly utilized. This requires preprocessing input data at the rate at which the accelerators can ingest and perform ML computations on the data. To avoid data stalls, the host CPU and RAM required for input data processing per accelerator core used for ML computations varies across jobs. Hence, the traditional approach of processing input data on ML accelerator hosts with a fixed hardware ratio leads to either under-utilizing the accelerators or the host CPU and RAM. In this paper, we address these concerns by building a disaggregated ML data processing system. We present tf.data service, an open-source disaggregated input data processing service built on top of tf.data in TensorFlow. We show that disaggregating data preprocessing has three key advantages for large-scale ML training jobs. First, the service can horizontally scale-out to right-size CPU/RAM host resources for data processing in each job, saving 32x training time and 26x cost, on average. Second, the service can share ephemeral preprocessed data results across jobs, to optimize CPU usage and reduce redundant computations. Finally, the service supports coordinated reads, a technique that avoids stragglers due to different input sizes in distributed training, reducing training time by 2.2x, on average. Our design is inspired by lessons learned from deploying tf.data service in production, including relaxing data visitation guarantees without impacting model accuracy.

cs.LG↗

arXiv2022

SHiFT: An Efficient, Flexible Search Engine for Transfer Learning

Transfer learning can be seen as a data- and compute-efficient alternative to training models from scratch. The emergence of rich model repositories, such as TensorFlow Hub, enables practitioners and researchers to unleash the potential of these models across a wide range of downstream tasks. As these repositories keep growing exponentially, efficiently selecting a good model for the task at hand becomes paramount. By carefully comparing various selection and search strategies, we realize that no single method outperforms the others, and hybrid or mixed strategies can be beneficial. Therefore, we propose SHiFT, the first downstream task-aware, flexible, and efficient model search engine for transfer learning. These properties are enabled by a custom query language SHiFT-QL together with a cost-based decision maker, which we empirically validate. Motivated by the iterative nature of machine learning development, we further support efficient incremental executions of our queries, which requires a careful implementation when jointly used with our optimizations.

cs.LG↗

arXiv2022

An Elastic Ephemeral Datastore using Cheap, Transient Cloud Resources

Spot instances are virtual machines offered at 60-90% lower cost that can be reclaimed at any time, with only a short warning period. Spot instances have already been used to significantly reduce the cost of processing workloads in the cloud. However, leveraging spot instances to reduce the cost of stateful cloud applications is much more challenging, as the sudden preemptions lead to data loss. In this work, we propose leveraging spot instances to decrease the cost of ephemeral data management in distributed data analytics applications. We specifically target ephemeral data as this large class of data in modern analytics workloads has low durability requirements; if lost, the data can be regenerated by re-executing compute tasks. We design an elastic, distributed ephemeral datastore that handles node preemptions transparently to user applications and minimizes data loss by redistributing data during node preemption warning periods. We implement our elastic datastore on top of the Apache Crail datastore and evaluate the system with various workloads and VM types. By leveraging spot instances, we show that we can run TPC-DS queries with 60\% lower cost compared to using on-demand VMs for the datastore, while only increasing end-to-end execution time by 2.1%.

cs.DC↗

arXiv2022

Plumber: Diagnosing and Removing Performance Bottlenecks in Machine Learning Data Pipelines

Input pipelines, which ingest and transform input data, are an essential part of training Machine Learning (ML) models. However, it is challenging to implement efficient input pipelines, as it requires reasoning about parallelism, asynchrony, and variability in fine-grained profiling information. Our analysis of over two million ML jobs in Google datacenters reveals that a significant fraction of model training jobs could benefit from faster input data pipelines. At the same time, our analysis indicates that most jobs do not saturate host hardware, pointing in the direction of software-based bottlenecks. Motivated by these findings, we propose Plumber, a tool for finding bottlenecks in ML input pipelines. Plumber uses an extensible and interpretable operational analysis analytical model to automatically tune parallelism, prefetching, and caching under host resource constraints. Across five representative ML pipelines, Plumber obtains speedups of up to 47x for misconfigured pipelines. By automating caching, Plumber obtains end-to-end speedups of over 50% compared to state-of-the-art tuners.

cs.LG↗

arXiv2022

Short-lived Datacenter

Serverless platforms have attracted attention due to their promise of elasticity, low cost, and fast deployment. Instead of using a fixed virtual machine (VM) infrastructure, which can incur considerable costs to operate and run, serverless platforms support short computations, triggered on demand, with cost proportional to fine-grain function execution time. However, serverless platforms offer a restricted execution environment. For example, functions have limited execution times, limited resources, and no support for networking between functions. In this paper, we explore what it takes to treat serverless platforms as short-lived, general purpose data-centers which can execute unmodified existing applications. As a first step in this quest, we have developed Boxer, a system providing an execution environment on top of existing functions-as-a-service platforms that allows users to seamlessly migrate conventional VM-based cloud services to serverless platforms. Boxer allows generic applications to benefit from the fine-grain elasticity of serverless platforms without having to modify applications to adopt a restrictive event-triggered programming model or orchestrate auxiliary systems for data communication. We implement Boxer on top of AWS Lambda and extend it to transparently provide standard network interfaces. We describe its implementation and demonstrate how it can be used to run off-the-shelf cloud applications with a degree of fine-grained elasticity not available on traditional VM-based platforms.

cs.DC↗

arXiv2021

How to use Persistent Memory in your Database

Persistent or Non Volatile Memory (PMEM or NVM) has recently become commercially available under several configurations with different purposes and goals. Despite the attention to the topic, we are not aware of a comprehensive empirical analysis of existing relational database engines under different PMEM configurations. Such a study is important to understand the performance implications of the various hardware configurations and how different DB engines can benefit from them. To this end, we analyze three different engines (PostgreSQL, MySQL, and SQLServer) under common workloads (TPC-C and TPC-H) with all possible PMEM configurations supported by Intel's Optane NVM devices (PMEM as persistent memory in AppDirect mode and PMEM as volatile memory in Memory mode). Our results paint a complex picture and are not always intuitive due to the many factors involved. Based on our findings, we provide insights on how the different engines behave with PMEM and which configurations and queries perform best. Our results show that using PMEM as persistent storage usually speeds up query execution, but with some caveats as the I/O path is not fully optimized. Additionally, using PMEM in Memory mode does not offer any performance advantage despite the larger volatile memory capacity. Through the extensive coverage of engines and parameters, we provide an important starting point for exploiting PMEM in databases and tuning relational engines to take advantage of this new technology.

cs.DB↗

arXiv2021

Modularis: Modular Relational Analytics over Heterogeneous Distributed Platforms

The enormous quantity of data produced every day together with advances in data analytics has led to a proliferation of data management and analysis systems. Typically, these systems are built around highly specialized monolithic operators optimized for the underlying hardware. While effective in the short term, such an approach makes the operators cumbersome to port and adapt, which is increasingly required due to the speed at which algorithms and hardware evolve. To address this limitation, we present Modularis, an execution layer for data analytics based on sub-operators, i.e.,composable building blocks resembling traditional database operators but at a finer granularity. To demonstrate the advantages of our approach, we use Modularis to build a distributed query processing system supporting relational queries running on an RDMA cluster, a serverless cloud platform, and a smart storage engine. Modularis requires minimal code changes to execute queries across these three diverse hardware platforms, showing that the sub-operator approach reduces the amount and complexity of the code. In fact, changes in the platform affect only sub-operators that depend on the underlying hardware. We show the end-to-end performance of Modularis by comparing it with a framework for SQL processing (Presto), a commercial cluster database (SingleStore), as well as Query-as-a-Service systems (Athena, BigQuery). Modularis outperforms all these systems, proving that the design and architectural advantages of a modular design can be achieved without degrading performance. We also compare Modularis with a hand-optimized implementation of a join for RDMA clusters. We show that Modularis has the advantage of being easily extensible to a wider range of join variants and group by queries, all of which are not supported in the hand-tuned join.

cs.DB↗

arXiv2021

Towards Demystifying Serverless Machine Learning Training

The appeal of serverless (FaaS) has triggered a growing interest on how to use it in data-intensive applications such as ETL, query processing, or machine learning (ML). Several systems exist for training large-scale ML models on top of serverless infrastructures (e.g., AWS Lambda) but with inconclusive results in terms of their performance and relative advantage over "serverful" infrastructures (IaaS). In this paper we present a systematic, comparative study of distributed ML training over FaaS and IaaS. We present a design space covering design choices such as optimization algorithms and synchronization protocols, and implement a platform, LambdaML, that enables a fair comparison between FaaS and IaaS. We present experimental results using LambdaML, and further develop an analytic model to capture cost/performance tradeoffs that must be considered when opting for a serverless infrastructure. Our results indicate that ML training pays off in serverless only for models with efficient (i.e., reduced) communication and that quickly converge. In general, FaaS can be much faster but it is never significantly cheaper than IaaS.

cs.DC↗

arXiv2021

tf.data: A Machine Learning Data Processing Framework

Training machine learning models requires feeding input data for models to ingest. Input pipelines for machine learning jobs are often challenging to implement efficiently as they require reading large volumes of data, applying complex transformations, and transferring data to hardware accelerators while overlapping computation and communication to achieve optimal performance. We present tf.data, a framework for building and executing efficient input pipelines for machine learning jobs. The tf.data API provides operators which can be parameterized with user-defined computation, composed, and reused across different machine learning domains. These abstractions allow users to focus on the application logic of data processing, while tf.data's runtime ensures that pipelines run efficiently. We demonstrate that input pipeline performance is critical to the end-to-end training time of state-of-the-art machine learning models. tf.data delivers the high performance required, while avoiding the need for manual tuning of performance knobs. We show that tf.data features, such as parallelism, caching, static optimizations, and non-deterministic execution are essential for high performance. Finally, we characterize machine learning input pipelines for millions of jobs that ran in Google's fleet, showing that input data processing is highly diverse and consumes a significant fraction of job resources. Our analysis motivates future research directions, such as sharing computation across jobs and pushing data projection to the storage layer.

cs.LG↗