Applications in data-parallel computing typically consist of multiple stages. In each stage, a set of intermediate parallel data flows (Coflow) is produced and transferred between servers to enable starting of next stage. While there has been much research on scheduling isolated coflows, the dependency between coflows in multi-stage jobs has been largely ignored. In this paper, we consider scheduling coflows of multi-stage jobs represented by general DAGs (Directed Acyclic Graphs) in a shared data center network, so as to minimize the total weighted completion time of jobs. This problem is significantly more challenging than the traditional coflow scheduling, as scheduling even a single multi-stage job to minimize its completion time is shown to be NP-hard.
In this paper, we propose a polynomial-time algorithm with approximation ratio of $O(μ\log(m)/\log(\log(m)))$, where $μ$ is the maximum number of coflows in a job and $m$ is the number of servers. For the special case that the jobs' underlying dependency graphs are rooted trees, we modify the algorithm and improve its approximation ratio. To verify the performance of our algorithms, we present simulation results using real traffic traces that show up to $53 \%$ improvement over the prior approach. We conclude the paper by providing a result concerning an optimality gap for scheduling coflows with general DAGs.
Datacenter networks routinely support the data transfers of distributed computing frameworks in the form of coflows, i.e., sets of concurrent flows related to a common task. The vast majority of the literature has focused on the problem of scheduling coflows for completion time minimization, i.e., to maximize the average rate at which coflows are dispatched in the network fabric. Modern applications, though, may generate coflows dedicated to online services and mission-critical computing tasks which have to comply with specific completion deadlines. In this paper, we introduce $\mathtt{DCoflow…
Many modern datacenter applications involve large-scale computations composed of multiple data flows that need to be completed over a shared set of distributed resources. Such a computation completes when all of its flows complete. A useful abstraction for modeling such scenarios is a {\em coflow}, which is a collection of flows (e.g., tasks, packets, data transmissions) that all share the same performance goal. In this paper, we present the first approximation algorithms for scheduling coflows over general network topologies with the objective of minimizing total weighted completion time. We …
The coflow scheduling problem has emerged as a popular abstraction in the last few years to study data communication problems within a data center. In this basic framework, each coflow has a set of communication demands and the goal is to schedule many coflows in a manner that minimizes the total weighted completion time. A coflow is said to complete when all its communication needs are met. This problem has been extremely well studied for the case of complete bipartite graphs that model a data center with full bisection bandwidth and several approximation algorithms and effective heuristics h…
The average coflow completion time (CCT) is the standard performance metric in coflow scheduling. However, standard CCT minimization may introduce unfairness between the data transfer phase of different computing jobs. Thus, while progress guarantees have been introduced in the literature to mitigate this fairness issue, the trade-off between fairness and efficiency of data transfer is hard to control. This paper introduces a fairness framework for coflow scheduling based on the concept of slowdown, i.e., the performance loss of a coflow compared to isolation. By controlling the slowdown it is…
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 simil…
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);…
Coflow is a recently proposed networking abstraction to help improve the communication performance of data-parallel computing jobs. In multi-stage jobs, each job consists of multiple coflows and is represented by a Directed Acyclic Graph (DAG). Efficiently scheduling coflows is critical to improve the data-parallel computing performance in data centers. Compared with hand-tuned scheduling heuristics, existing work DeepWeave [1] utilizes Reinforcement Learning (RL) framework to generate highly-efficient coflow scheduling policies automatically. It employs a graph neural network (GNN) to encode …
Cloud computing technology has been one of the most critical developments in provisioning both hardware and software infrastructure in recent years. Container technology is a new cloud technology that boosts the booting of applications, increases the ability to deploy applications on containers and improves the host machine resource sharing. Thus, enhancing a cloud container system needs a robust algorithm that deploys the applications efficiently. Most of the schedulers associated with container technology are focused on load balancing for increasing container performance. The traffic over ne…
State Machine Replication (SMR) is a fundamental approach to designing service with fault tolerance. However, its requirement for the deterministic execution of transactions often results in single-threaded replicas, which cannot fully exploit the multicore capabilities of today's processors. Therefore, parallel SMR has become a hot topic of recent research. The basic idea behind it is that independent transactions can be executed in parallel, while dependent transactions must be executed in their relative order to ensure consistency among replicas. The dependency detection of existing paralle…
In the federated scheduling approaches in multiprocessor systems, a task either 1) is restricted to execute sequentially on a single processor or 2) has exclusive access to the assigned processors. There have been several positive results to conduct good federated scheduling policies, which have constant speedup factors with respect to any optimal federated scheduling algorithm. This paper answers an open question: "For constrained-deadline task systems with directed acyclic graph (DAG) dependency structures, do federated scheduling policies have a constant speedup factor with respect to any o…
Task-based programming models have become very popular, as they offer an attractive solution to parallelize serial application code with task and data annotations. They usually depend on a runtime system that schedules the tasks to multiple cores in parallel while resolving any data hazards. However, existing runtime system implementations are not ready to scale well on emerging manycore processors, as they often rely on centralized structures and/or locks on shared structures in a cache-coherent memory. We propose design choices, policies and mechanisms to enhance runtime system scalability f…
For a cloud service provider, delivering optimal system performance while fulfilling Quality of Service (QoS) obligations is critical for maintaining a viably profitable business. This goal is often hard to attain given the irregular nature of cloud computing jobs. These jobs expect high QoS on an on-demand fashion, that is on random arrival. To optimize the response to such client demands, cloud service providers organize the cloud computing environment as a multi-tier architecture. Each tier executes its designated tasks and passes the job to the next tier; in a fashion similar, but not iden…
The containerized services allocated in the mobile edge clouds bring up the opportunity for large-scale and real-time applications to have low latency responses. Meanwhile, live container migration is introduced to support dynamic resource management and users' mobility. However, with the expansion of network topology scale and increasing migration requests, the current multiple migration planning and scheduling algorithms of cloud data centers can not suit large-scale scenarios in edge computing. The user mobility-induced live migrations in edge computing require near real-time level scheduli…
The University of Sydney（悉尼大学）；CSIRO Data61（澳大利亚联邦科学与工业研究组织数据61所）
Dependency-aware job scheduling in the cluster is NP-hard. Recent work shows that Deep Reinforcement Learning (DRL) is capable of solving it. It is difficult for the administrator to understand the DRL-based policy even though it achieves remarkable performance gain. Therefore the complex model-based scheduler is not easy to gain trust in the system where simplicity is favored. In this paper, we give the multi-level explanation framework to interpret the policy of DRL-based scheduling. We dissect its decision-making process to job level and task level and approximate each level with interpreta…
We consider the classical problem of scheduling task graphs corresponding to complex applications on distributed computing systems. A number of heuristics have been previously proposed to optimize task scheduling with respect to metrics such as makespan and throughput. However, they tend to be slow to run, particularly for larger problem instances, limiting their applicability in more dynamic systems. Motivated by the goal of solving these problems more rapidly, we propose, for the first time, a graph convolutional network-based scheduler (GCNScheduler). By carefully integrating an inter-task …
GPUs are readily available in cloud computing and personal devices, but their use for data processing acceleration has been slowed down by their limited integration with common programming languages such as Python or Java. Moreover, using GPUs to their full capabilities requires expert knowledge of asynchronous programming. In this work, we present a novel GPU run time scheduler for multi-task GPU computations that transparently provides asynchronous execution, space-sharing, and transfer-computation overlap without requiring in advance any information about the program dependency structure. W…
This paper presents a multiagent approach as a paradigm for scheduling parallel jobs in a parallel system. Scheduling parallel jobs is performed as a means to balance the load of a system in order to improve the performance of a parallel application. Parallel job scheduling is presented as a mapping between two graphs: one represents the dependency of jobs and the other represents the interconnection among processors. The usual implementation of parallel job scheduling algorithms is via the master-slave paradigm. The master-slave paradigm has inherent communication bottleneck that reduces the …
Scheduling job flows efficiently and rapidly on distributed computing clusters is one of huge challenges for daily operation of data centers. In a practical scenario, a single job consists of numerous stages with complex dependency relation represented as a Directed Acyclic Graph (DAG) structure. Nowadays a data center usually equips with a cluster of heterogeneous computing servers which are different in the hardware/software configuration. From both the cost saving and environmental friendliness, the data centers could benefit a lot from optimizing the job scheduling problems in the heteroge…
HyperParallel-MoE: Multi-Core Interleaved Scheduling for Fast MoE Training on Ascend NPUs
University of Science and Technology of China（中国科学技术大学）；Huawei Technologies Co., Ltd（华为技术有限公司）；Peking University（北京大学）；Institute of Artificial Intelligence, Hefei Comprehensive National Science Center（合肥综合性国家科学中心人工智能研究院）
Many scientific workflows can be represented by a Directed Acyclic Graph (DAG) where each node represents a task, and there will be a directed edge between two tasks if and only if there is a dependency relationship between the two i.e. the second one can not be started unless the first one is finished. Due to the increasing computational requirements of these workflows, they are deployed on cloud computing systems. Scheduling of workflows on such systems to achieve certain goals(e.g. minimization of makespan, cost, or maximization of reliability, etc.) remains an active area of research. In t…
Many scientific workflows can be modeled as a Directed Acyclic Graph (henceforth mentioned as DAG) where the nodes represent individual tasks and the directed edges represent data and control flow dependency between two tasks. Due to large computational resource requirements, a single cloud cannot meet the requirements of the workflow. Hence, a multi-cloud system, where multiple cloud providers pool their resources together becomes a good solution. The major objectives considered while scheduling the tasks present in a task graph include execution cost and makespan. In this paper, we present C…
