Large-scale datacenters have been widely used to host cloud services, which are typically allocated to different virtual machines (VMs) through resource multiplexing across shared physical servers. Although recent studies have primarily focused on harnessing live migration of VMs to achieve load balancing and power saving among different servers, there has been little attention on the incurred performance interference and cost on both source and destination servers during and after such VM migration. To avoid potential violations of service-level-agreement (SLA) demanded by cloud applications, this paper proposes iAware, a lightweight interference-aware VM live migration strategy. It empirically captures the essential relationships between VM performance interference and key factors that are practically accessible through realistic experiments of benchmark workloads on a Xen virtualized cluster platform. iAware jointly estimates and minimizes both migration and co-location interference among VMs, by designing a simple multi-resource demand-supply model. Extensive experiments and complementary large-scale simulations are conducted to validate the performance gain and runtime overhead of iAware in terms of I/O and network throughput, CPU consumption, and scalability, compared to the traditional interference-unaware VM migration approaches. Moreover, we demonstrate that iAware is flexible enough to cooperate with existing VM scheduling or consolidation policies in a complementary manner, such that the load balancing or power saving can still be achieved without sacrificing performance.

http://grid.hust.edu.cn/fmliu/tc-iaware.pdf

iAware: Making Live Migration of Virtual Machines Interference-Aware in the Cloud

We analyze the manycore scalability of five widely-deployed file systems, namely, ext4, XFS, btrfs, F2FS, and tmpfs, by using our open source benchmark suite, FXMARK. FXMARK implements 19 microbenchmarks to stress specific components of each file system and includes three application benchmarks to measure the macroscopic scalability behavior. We observe that file systems are hidden scalability bottlenecks in many I/O-intensive applications even when there is no apparent contention at the application level. We found 25 scalability bottlenecks in file systems, many of which are unexpected or counterintuitive. We draw a set of observations on file system scalability behavior and unveil several core aspects of file system design that systems researchers must address.

/pdf/understanding-manycore-scalability-of-file-systems-4g66xoovlg.pdf

Understanding manycore scalability of file systems

Virtual machine (VM) consolidation in Cloud computing provides a great opportunity for energy saving. However, the obligation of providing suitable quality of service to end users leads to the necessity in dealing with energy-performance tradeoff. In this paper, we propose a redesigned energy-aware heuristic framework for VM consolidation to achieve a better energy-performance tradeoff. There are two main contributions in the framework: (1) establish a service level agreement (SLA) violation decision algorithm to decide whether a host is overload with SLA violation; (2) minimum power and maximum utilization policy is then proposed to improve the Minimum Power policy in previous work. Finally, we have evaluated our framework through simulation on large-scale experiments driven by workload traces from more than a thousand VMs, and the results show that our framework outperforms previous work. Specifically, it guarantees 21---34 % decrease in energy consumption, 84---92 % decrease in SLA violation, 87---94 % decrease in energy-performance metric, and 63 % decrease in execution time. And we further discuss why the redesigned framework outperforms the previous design.

An energy-aware heuristic framework for virtual machine consolidation in Cloud computing

Considered as the next generation computing model, cloud computing plays an important role in scientific and commercial computing area and draws great attention from both academia and industry fields. Under cloud computing environment, data center consist of a large amount of computers, usually up to millions, and stores petabyte even exabyte of data, which may easily lead to the failure of the computers or data. The large amount of computers composition not only leads to great challenges to the scalability of the data center and its storage system, but also results in high hardware infrastructure cost and power cost. Therefore, fault-tolerance, scalability, and power consumption of the distributed storage for a data center becomes key part in the technology of cloud computing, in order to ensure the data availability and reliability. In this paper, a survey is made on the state of art of the key technologies in cloud computing in the following aspects: Design of data center network, organization and arrangement of data, strategies to improve fault-tolerance, methods to save storage space, and energy. Firstly, many kinds of classical topologies of data center network are introduced and compared. Secondly, kinds of current fault-tolerant storage techniques are discussed, and data replication and erasure code strategies are especially compared. Thirdly, the main current energy saving technology is addressed and analyzed. Finally, challenges in distributed storage are reviewed as well as future research trends are predicted.

/pdf/key-technologies-of-distributed-storage-for-cloud-computing-4ouq7mgjth.pdf

Key Technologies of Distributed Storage for Cloud Computing

Fast storage devices are an emerging solution to satisfy data-intensive applications. They provide high transaction rates for DBMS, low response times for Web servers, instant on-demand paging for applications with large memory footprints, and many similar advantages for performance-hungry applications. In spite of the benefits promised by fast hardware, modern operating systems are not yet structured to take advantage of the hardware’s full potential. The software overhead caused by an OS, negligible in the past, adversely impacts application performance, lessening the advantage of using such hardware. Our analysis demonstrates that the overheads from the traditional storage-stack design are significant and cannot easily be overcome without modifying the hardware interface and adding new capabilities to the operating system.In this article, we propose six optimizations that enable an OS to fully exploit the performance characteristics of fast storage devices. With the support of new hardware interfaces, our optimizations minimize per-request latency by streamlining the I/O path and amortize per-request latency by maximizing parallelism inside the device. We demonstrate the impact on application performance through well-known storage benchmarks run against a Linux kernel with a customized SSD. We find that eliminating context switches in the I/O path decreases the software overhead of an I/O request from 20 microseconds to 5 microseconds and a new request merge scheme called Temporal Merge enables the OS to achieve 87p to 100p of peak device performance, regardless of request access patterns or types. Although the performance improvement by these optimizations on a standard SATA-based SSD is marginal (because of its limited interface and relatively high response times), our sensitivity analysis suggests that future SSDs with lower response times will benefit from these changes. The effectiveness of our optimizations encourages discussion between the OS community and storage vendors about future device interfaces for fast storage devices.

/pdf/optimizing-the-block-i-o-subsystem-for-fast-storage-devices-2pvrbt112q.pdf

Optimizing the Block I/O Subsystem for Fast Storage Devices

Consolidation and scaling down can be used as enabling techniques to effectively manage power for distributed systems. To support power management for cloud computing with consolidation and scaling down, it is necessary to study the effects of mixed workloads and the consolidation on the performance of application in the virtual machines.Accordingly, we present two studies investigating the effects of VM co-location and CPU thermal management along with the performance in this paper. We conducted experiments to show the effects of performance interference when VMs with different characteristics are consolidated in a single server and analyzed the results so that we can use the results to design a workload-aware VM live scheduler with distributed power management for cloud computing.

/pdf/enabling-consolidation-and-scaling-down-to-provide-power-4nn01cg3h8.pdf

Enabling consolidation and scaling down to provide power management for cloud computing

Emerging non-volatile memory technologies as a disk drive replacement raise some issues of software stack and interfaces, which have not been considered in disk-based storage systems. In this work, we present new cooperative schemes including software and hardware to address performance issues with deploying storage-class memory technologies as a storage device. First, we propose a new polling scheme called dynamic interval polling to avoid the unnecessary polls and reduce the burden on storage system bus. Second, we propose a pipelined execution between storage device and host OS called pipelined post I/O processing. By extending vendor-specific I/O interfaces between software and hardware, we can improve the responsiveness of I/O requests with no sacrifice of throughput.

/pdf/dynamic-interval-polling-and-pipelined-post-i-o-processing-2jzogyep5l.pdf

Dynamic interval polling and pipelined post I/O processing for low-latency storage class memory

Energy efficiency in cloud computing is becoming more and more important for IT operators of data centers. Several effort to use low power machines in the data center level has been explored. Also, data processing frameworks such as MapReduce and Hadoop are frequently used to process data intensive jobs. However, there have not been an extensive study on the impact of low power computers on such data processing frameworks. Actually, development of low power computers is demanding the architectural paradigm shift for cloud applications. In this paper, we evaluate Apache Hadoop on low power machines and study the feasibility of them in cloud systems. We also propose AnSwer (Augmentation and Substitution), an energy saving method to reduce energy consumption by introducing low power machines. In An-Swer, augmentation and substitution complement each other to prevent data loss and to improve overall power consumption.

/pdf/towards-energy-proportional-cloud-for-data-processing-3igtfemx5n.pdf

Towards energy proportional cloud for data processing frameworks

Write-through caching in modern disk drives enables the protection of data in the event of power failures as well as from certain disk errors when the write-back cache does not. Host system can achieve these benefits at the price of significant performance degradation, especially for small disk writes. We present new block-level techniques to address the performance problem of write-through caching disks. Our techniques are strongly motivated by some interesting results when the disk-level caching is turned off. By extending the conventional request merging, request bridging increases the request size and amortizes the inherent delays in the disk drive across more bytes of data. Like sector interleaving, request interleaving rearranges requests to prevent the disk head from missing the target sector position in close proximity, and thus reduces disk latency. We have evaluated our block-level approach using a variety of I/O workloads and shown that it increases disk I/O throughput by up to about 50%. For some real-world workloads, the disk performance is comparable or even superior to that of using the write-back disk cache. In practice, our simple yet effective solutions achieve better tradeoffs between data reliability and disk performance when applied to write-through caching disks.

Request Bridging and Interleaving: Improving the Performance of Small Synchronous Updates under Seek-Optimizing Disk Subsystems

Energy efficiency is becoming a key issue for IT operators and data centres. To provide Power Save Mode (PSM) for data processing applications in such environments, power down method as a scaling down technique is an effective solution toward energy proportionality. The existing solutions for the PSM are inefficient in the overall energy management, taking the full replication approach. We propose an efficient replica redistribution algorithm, and our experimental results show that our system significantly reduces the network usage and the elapsed time. Our system leads to a slight increase in running time compared with the full replication approach.

Hyeong S. Kim

Papers

Enabling consolidation and scaling down to provide power management for cloud computing

Dynamic interval polling and pipelined post I/O processing for low-latency storage class memory

Towards energy proportional cloud for data processing frameworks

Request Bridging and Interleaving: Improving the Performance of Small Synchronous Updates under Seek-Optimizing Disk Subsystems

Systematic approach of using power save mode for cloud data processing services