Performance modeling can enhance the resource efficiency of clusters and distributed data processing applications, however, the modeling data available is frequently insufficient. Simultaneously, workloads regularly exhibit similar traits that could be leveraged to refine and enhance performance models for greater resilience. Numerous recent studies have made efforts in this direction by assessing the similarity of workloads in one way or another, although the effectiveness of the similarity assessment techniques employed for data exploitation remains questionable.This paper examines such techniques, investigating their underlying presumptions and common strategies for assessing similarity. We carry out a range of experiments in a public cloud environment using representative algorithms, datasets, and parameterizations to thoroughly showcase both the strengths and weaknesses of current methods in the discipline. This analysis serves as a basis to propose ideas for future research directions on workload similarity, with the ultimate goal of deriving informed resource configurations for big data analytics workloads.

Distributed dataflow systems like Spark and Flink enable data-parallel processing of large datasets on clusters. Yet, selecting appropriate computational resources for dataflow jobs is often challenging. For efficient execution, individual resource allocations, such as memory and CPU cores, must meet the specific resource requirements of the job. An alternative to selecting a static resource allocation for a job execution is autoscaling as implemented for example by Spark.In this paper, we evaluate the resource efficiency of autoscaling batch data processing jobs based on resource demand both conceptually and experimentally by analyzing a new dataset of Spark job executions on Google Dataproc Serverless. In our experimental evaluation, we show that there is no significant resource efficiency gain over static resource allocations. We found that the inherent conceptual limitations of such autoscaling approaches are the inelasticity of node size as well as the inelasticity of the ratio of memory to CPU cores.

Performance modeling can help to improve the resource efficiency of clusters and distributed dataflow applications, yet the available modeling data is often limited. Collaborative approaches to performance modeling, characterized by the sharing of performance data or models, have been shown to improve resource efficiency, but there has been little focus on actual data sharing strategies and implementation in production environments. This missing building block holds back the realization of proposed collaborative solutions.In this paper, we envision, design, and evaluate a peer-to-peer performance data sharing approach for collaborative performance modeling of distributed dataflow applications. Our proposed data distribution layer enables access to performance data in a decentralized manner, thereby facilitating collaborative modeling approaches and allowing for improved prediction capabilities and hence increased resource efficiency. In our evaluation, we assess our approach with regard to deployment, data replication, and data validation, through experiments with a prototype implementation and simulation, demonstrating feasibility and allowing discussion of potential limitations and next steps.

Distributed dataflow systems such as Apache Spark or Apache Flink enable parallel, in-memory data processing on large clusters of commodity hardware. Consequently, the appropriate amount of memory to allocate to the cluster is a crucial consideration.

Choosing a good resource configuration for big data analytics applications can be challenging, especially in cloud environments. Automated approaches are desirable as poor decisions can reduce performance and raise costs. The majority of existing automated approaches either build performance models from previous workload executions or conduct iterative resource configuration profiling until a near-optimal solution has been found. In doing so, they only obtain an implicit understanding of the underlying infrastructure, which is difficult to transfer to alternative infrastructures and, thus, profiling and modeling insights are not sustained beyond very specific situations. We present Perona, a novel approach to robust infrastructure fingerprinting for usage in the context of big data analytics. Perona employs common sets and configurations of benchmarking tools for target resources, so that resulting benchmark metrics are directly comparable and ranking is enabled. Insignificant benchmark metrics a red is carded by learning a low-dimensional representation of the input metric vector, and previous benchmark executions are taken into consideration for context-awareness as well, allowing to detect resource degradation. We evaluate our approach both on data gathered from our own experiments as well as within related works for resource configuration optimization, demonstrating that Perona captures the characteristics from benchmark runs in a compact manner and produces representations that can be used directly.

Selecting appropriate computational resources for data processing jobs on large clusters is difficult, even for expert users like data engineers. Inadequate choices can result in vastly increased costs, without significantly improving performance. One crucial aspect of selecting an efficient resource configuration is avoiding memory bottlenecks. By knowing the required memory of a job in advance, the search space for an optimal resource configuration can be greatly reduced.Therefore, we present Ruya, a method for memory-aware optimization of data processing cluster configurations based on iteratively exploring a narrowed-down search space. First, we perform job profiling runs with small samples of the dataset on just a single machine to model the job’s memory usage patterns. Second, we prioritize cluster configurations with a suitable amount of total memory and within this reduced search space, we iteratively search for the best cluster configuration with Bayesian optimization. This search process stops once it converges on a configuration that is believed to be optimal for the given job. In our evaluation on a dataset with 1031 Spark and Hadoop jobs, we see a reduction of search iterations to find an optimal configuration by around half, compared to the baseline.

Many scientific workflow scheduling algorithms need to be informed about task runtimes a-priori to conduct efficient scheduling. In heterogeneous cluster infrastructures, this problem becomes aggravated because these runtimes are required for each task-node pair. Using historical data is often not feasible as logs are typically not retained indefinitely and workloads as well as infrastructure changes. In contrast, online methods, which predict task runtimes on specific nodes while the workflow is running, have to cope with the lack of example runs, especially during the start-up.

Scientific workflow management systems like Nextflow support large-scale data analysis by abstracting away the details of scientific workflows. In these systems, workflows consist of several abstract tasks, of which instances are run in parallel and transform input partitions into output partitions. Resource managers like Kubernetes execute such workflow tasks on cluster infrastructures. However, these resource managers only consider the number of CPUs and the amount of available memory when assigning tasks to resources; they do not consider hardware differences beyond these numbers, while computational speed and memory access rates can differ significantly.We propose Tarema, a system for allocating task instances to heterogeneous cluster resources during the execution of scalable scientific workflows. First, Tarema profiles the available infrastructure with a set of benchmark programs and groups cluster nodes with similar performance. Second, Tarema uses online monitoring data of tasks, assigning labels to tasks depending on their resource usage. Third, Tarema uses the node groups and task labels to dynamically assign task instances evenly to resources based on resource demand. Our evaluation of a prototype implementation for Kubernetes, using five real-world Nextflow workflows from the popular nf-core framework and two 15-node clusters consisting of different virtual machines, shows a mean reduction of isolated job runtimes by 19.8% compared to popular schedulers in widely-used resource managers and 4.54% compared to the heuristic SJFN, while providing a better cluster usage. Moreover, executing two long-running workflows in parallel and on restricted resources shows that Tarema is able to reduce the runtimes even more while providing a fair cluster usage.

With the growing amount of data, data processing workloads and the management of their resource usage becomes increasingly important. Since managing a dedicated infrastructure is in many situations infeasible or uneconomical, users progressively execute their respective workloads in the cloud. As the configuration of workloads and resources is often challenging, various methods have been proposed that either quickly profile towards a good configuration or determine one based on data from previous runs. Still, performance data to train such methods is often lacking and must be costly collected.In this paper, we propose a collaborative approach for sharing anonymized workload execution traces among users, mining them for general patterns, and exploiting clusters of historical workloads for future optimizations. We evaluate our prototype implementation for mining workload execution graphs on a publicly available trace dataset and demonstrate the predictive value of workload clusters determined through traces only.

Distributed dataflow systems like Apache Flink and Apache Spark simplify processing large amounts of data on clusters in a data-parallel manner. However, choosing suitable cluster resources for distributed dataflow jobs in both type and number is difficult, especially for users who do not have access to previous performance metrics. One approach to overcoming this issue is to have users share runtime metrics to train context-aware performance models that help find a suitable configuration for the job at hand. A problem when sharing runtime data instead of trained models or model parameters is that the data size can grow substantially over time.This paper examines several clustering techniques to minimize training data size while keeping the associated performance models accurate. Our results indicate that efficiency gains in data transfer, storage, and model training can be achieved through training data reduction. In the evaluation of our solution on a dataset of runtime data from 930 unique distributed dataflow jobs, we observed that, on average, a 75% data reduction only increases prediction errors by one percentage point.

Analyzing large datasets with distributed dataflow systems requires the use of clusters. Public cloud providers offer a large variety and quantity of resources that can be used for such clusters. However, picking the appropriate resources in both type and number can often be challenging, as the selected configuration needs to match a distributed dataflow job's resource demands and access patterns. A good cluster configuration avoids hardware bottlenecks and maximizes resource utilization, avoiding costly overprovisioning.We propose a collaborative approach for finding optimal cluster configurations based on sharing and learning from historical runtime data of distributed dataflow jobs. Collaboratively shared data can be utilized to predict runtimes of future job executions through the use of specialized regression models. However, training prediction models on historical runtime data that were produced by different users and in diverse contexts requires the models to take these contexts into account.