Eventually-consistent federated scheduling for data center workloads

Spatio-temporal thermal-aware job scheduling to minimize energy consumption in virtualized heterogeneous data centers

Enumerating consistent global states of a computation is a fundamental problem in parallel computing with applications to debugging, testing and runtime verification of parallel programs. Breadth-first search (BFS) enumeration is especially useful for these applications as it finds an erroneous consistent global state with the least number of events possible. The total number of executed events in a global state is called its rank. BFS also allows enumeration of all global states of a given rank or within a range of ranks. If a computation on n processes has m events per process on average, then the traditional BFS (Cooper-Marzullo and its variants) requires \(\mathcal {O}(\frac{m^{n-1}}{n})\) space in the worst case, whereas our algorithm performs the BFS requires \(\mathcal {O}(m^2n^2)\) space. Thus, we reduce the space complexity for BFS enumeration of consistent global states exponentially, and give the first polynomial space algorithm for this task. In our experimental evaluation of seven benchmarks, traditional BFS fails in many cases by exhausting the 2 GB heap space allowed to the JVM. In contrast, our implementation uses less than 60 MB memory and is also faster in many cases.

As the demand for large scale computing is rapidly increasing to serve billions of users across the world, more powerful and densely packed server configurations are being used. Often in developing countries, and in small and medium enterprises, it is hard to place such servers in sound-proof server rooms. Hence, servers are typically placed in close proximity to employees. The noise from the cooling fans in servers adversely affects employees' health, and reduces their productivity. In this paper, we provide a framework for computer architects to measure the acoustic profile in a data center along with the temperature profile, and estimate the sound power levels at points of interest. Additionally, we studied the noise levels obtained upon using algorithms targeted at homogenizing the temperature profile. We found that these algorithms result in high noise levels, sometimes above the permissible levels. So, we propose two heuristics to redistribute workloads in a data center such that noise can be reduced at certain target locations. We obtain a noise reduction of 2-13 dB when compared with uniform workload distribution, and upto 16 dB as compared to temperature aware workload placement, with a reduction of at least 5-6 dB in 75% of the cases. The performance overhead is limited to 1%.

In distributed computing frameworks like MapReduce, Spark, and Dyrad, a coflow is a set of flows transferring data between two stages of a job. The job cannot start its next stage unless all flows in the coflow finish. To improve the execution performance of such a job, it is crucial to reduce the completion time of a coflow, as it can contribute more than 50 percent of the job completion time. While several coflow schedulers have been proposed, we observe that routing, as a factor greatly impacting the Coflow Completion Time (CCT), has not been well considered. In this article, we focus on the coflow scheduling problem and jointly consider routing and bandwidth allocation. We begin by providing an analytical solution to the problem of optimal bandwidth allocation with pre-determined routes. In the following, we formulate the problem of scheduling a single coflow as a Non-linear Mixed Integer Programming problem and present its relaxed convex optimization problem. We further propose two algorithms, CoRBA and its simplified version: CoRBA-fast that solve the single coflow scheduling problem with a joint consideration of routing and bandwidth allocation. Lastly, to address multiple coflows in online scheduling, we propose an online scheduler named OnCoRBA. By comparing with the start-of-the-art algorithms and schedulers via simulations, we demonstrate that CoRBA and CoRBA-fast reduce the CCT by 30-400 percent and the OnCoRBA scheduler reduces the average online CCT by 20-230 percent. In addition, CoRBA-fast can be hundreds times faster than CoRBA with around 8 percent performance degradation compared to CoRBA, which makes the use of CoRBA-fast very appropriate in practice.

In this paper we analyze the problem of optimal task scheduling for data centers. Given the available resources and tasks, we propose a fast distributed iterative algorithm, which operates over a large scale network of nodes, and allows each of the interconnected nodes to reach agreement to an optimal solution in a finite number of time steps. More specifically, the algorithm (i) is guaranteed to converge to the exact optimal scheduling plan in a finite number of time steps and, (ii) once the goal of task scheduling is achieved, it exhibits distributed stopping capabilities (i.e., it allows the nodes to distributely determine whether they can terminate the operation of the algorithm). Furthermore, the proposed algorithm operates exclusively with quantized values (i.e., the information stored, processed and exchanged between neighboring agents is subject to deterministic uniform quantization) and relies on event-driven updates (e.g., to reduce energy consumption, communication bandwidth, network congestion, and/or processor usage). We also provide examples to illustrate the operation, performance, and potential advantages of the proposed algorithm. Finally, by using extensive empirical evaluations through simulations we show that the operation of our proposed algorithm is suitable for large scale networks such as data centers.

Communication in data centers often involves many parallel flows that all share the same performance goal. A useful abstraction, coflow, is proposed to express the communication requirements of prevalent data parallel paradigms. The multiple coflow routing and scheduling problem faces challenges when deriving a good theoretical performance ratio because coexisting coflows will compete for the same network resources such as link bandwidths. In this paper, we focus on the coflow problem in the most popular data center infrastructure: the Leaf-Spine topology. We first formulate the problem and study the path selection issue on this two-tier structure. In order to minimize the average coflow completion time (CCT), we propose a Multi- hop Coflow Routing and Scheduling strategy (MCRS) for inter-coflows and intra-coflows and prove that our method has a reasonably good competitive ratio. Extensive experiments show that MCRS outperforms other state-of- art schemes.

ABSTRACTTask scheduling algorithms for data centers must be capable of making instantaneous decisions based on the current state of the system. However, due to information limitations, these scheduling algorithms often fail to achieve optimal scheduling plans. To address the information bottleneck faced in DAG (Directed Acyclic Graph) task scheduling within data centers, this paper proposes a deep reinforcement learning scheduling model based on a DAG attention mechanism. This model utilizes the attention mechanism to capture the potential relationships between dependent tasks, thereby improving scheduling efficiency and system performance under limited information conditions. The experimental results indicate that our DAG attention mechanism can significantly reduce makespan.

Greening cloud data centers in an economical way by energy trading with power grid

Existing flow scheduling schemes in Data Center Network (DCN) are designed mainly to minimize the flow complete time (FCT) of short flows and do not consider optimizing the FCT of latency-sensitive long flows (e.g. VR video streaming, interactive artificial intelligence questiona and the information-agnostic scheme (i.e. PIAS), which is based on the premise that flow size is not known a priori, requires a central server, causing a poor scalability in large network scales. Given the limitations of existing solutions, in this paper, we propose a distributed information-agnostic flow scheduling scheme (DIAS), which minimizes the FCT of both short flows and latency-sensitive long flows. In DIAS, packets are forwarded complying with their priorities, which are determined based on packets wait-time that is defined as staying time in end hosts' send buffers, and the longer a packet stays in a send buffer, the lower its priority. Meanwhile, instead of utilizing a central server to collect traffic load information, each switch feeds traffic load information which is used to adjust the thresholds of determining packets priority back to end hosts via ACK packets. The experimental results in ns-3 simulator show that DIAS reduces FCT by up to 54.7% and 50.1 % over DCTCP and $L^{2}\mathbf{DCT}$ , respectively. Besides, DIAS ensures a smaller FCT of latency-sensitive long flows and performs better than PIAS.

This chapter contains sections titled: Introduction Related Work Intermachine Scheduling Intramachine Scheduling Evaluation Conclusion References

Wide-area distributed storage system relies on the data replication to achieve the reliability. One of the main challenges brought by data replication is the consistency. As the variety of applications increases, the level of data consistency of the system has evolved from a single level of consistency, such as traditional strong consistency and eventual consistency, to the best level of consistency by adaptively selecting based on the application. However, the existing adaptive consistency schemes are mostly modified based upon the traditional eventual consistency protocol such as Multi-Paxos and rarely consider the diversity of the Data Centers (DCs), which results in high latency and low throughput with limited bandwidth. In this paper, we propose Grouping Multi-Paxos (GM-Paxos), an adaptive consistency protocol based on Multi-Paxos to alleviate the limitation of bandwidth in the system. Taking the characteristics of the DCs and the system environment into consideration, we design a grouping algorithm to find the primary group and only send the request content to the DCs within the primary group at first, in order to save the bandwidth consumption. Experiments show that GM-Paxos can significantly reduce the latency and improve the throughput. Besides, we synchronize the DCs which have high real-time data requirements at first to ensure the reliability of the system.

Modern data centers typically contain thousands of servers, providing various computing and storage services for users. The strategy to provide reliable and high-performance online services is to over-allocate resources for online services, which results in a waste of cluster resources. Therefore, cloud vendors tend to co-locate online services and offline batch jobs into the same cluster to improve resource utilization. However, the co-location leads to contention on shared resources and causes mutual performance interference, which may degrade the QoS (Quality of Service) of online services. We present a performance interference model based on linear regression to predict the performance interference. Furthermore, the model can perceive the status of servers in real-time for more refined and accurate prediction. Then, we design an interference-aware workload scheduling strategy that can schedule batch jobs to the server while introducing minimal interference. The evaluation demonstrates that our scheduling strategy can at best increase the throughput of batch jobs by 48.95% and 27.09% compared with round-robin scheduling and random scheduling while guaranteeing the QoS of online services. The paper aims to provide some elements of answering the following general question: How do we design cloud systems and data centers scheduling strategies that can withstand the human pressure of global-scale use and still provide robust and secure services to end-users?KeywordsCloud computingWorkload co-locationPerformance interferencePerformance predictionWorkload schedulingData centers

Hybrid ant genetic algorithm for efficient task scheduling in cloud data centers

Eventual consistency is demanded nowadays in geo-replicated services that need to be highly scalable and available. According to the CAP constraints, when network partitions may arise, a distributed service should choose between being strongly consistent or being highly available. Since scalable services should be available, a relaxed consistency (while the network is partitioned) is the preferred choice. Eventual consistency is not a common data-centric consistency model, but only a state convergence condition to be added to a relaxed consistency model. There are still several aspects of eventual consistency that have not been analysed in depth in previous works: 1. which are the oldest replication proposals providing eventual consistency, 2. which replica consistency models provide the best basis for building eventually consistent services, 3. which mechanisms should be considered for implementing an eventually consistent service, and 4. which are the best combinations of those mechanisms for achieving different concrete goals. This paper provides some notes on these important topics.

Existing data centers are each individually owned and operated by a single entity. This situation creates an excessive financial burden upon each entity through the need to overprovision for hardware. While this state of affairs enables a secure and an efficient maintenance scheme for the data center, the financial drawbacks are perhaps excessive. To address the concern, we propose the idea of a cooperative virtual data center among multiple entities that is founded on the principal of fair resource sharing amongst the entities.