Search

In Infrastructure-as-a-Service clouds, virtual machine provisioning time is an important metric for evaluating the quality of service (QoS) provided by the service providers. VM provisioning is usually time-consuming, especially when many VMs boot up simultaneously. We call such phenomenons "Boot Storm". In practice, many VMs would run on a single compute node for higher resource utilization. This multiple VMs and single compute node scenario easily encounters Boot Storm. Boot Storm leads to extremely high latency of VM provisioning and badly affects user's experience. In this paper, we propose VMBKS, which is a cloud shared memory cache system based on booting kernel. We exploit VMs' correlations to organize VM image files, to reduce cache size and mitigate the effect of Boot Storm. We use booting kernel to construct cache and share the booting kernel to correlative VMs. Evaluation shows that VMBKS can speed up VM provisioning time by up to 60 % and mitigate the effect of Boot Storm significantly. [ABSTRACT FROM AUTHOR]

Parameter-server frameworks play an important role in scaling-up distributed deep learning algorithms. However, the constant growth of neural network size has led to a serious bottleneck on exchanging parameters across machines. Recent efforts rely on manually setting a parameter-exchanging interval to reduce communication overhead, regardless of the parameter-server's resource availability as well. It may face poor performance or inaccurate results for inappropriate interval. Meanwhile, request burst may occur, exacerbating the bottleneck. In this paper, we propose an approach to automatically set the optimal exchanging interval, aiming to remove the parameter-exchanging bottleneck and to evenly utilize resources without losing training accuracy. The key idea is to increase the interval on different training nodes on the basis of the knowledge of available resources and choose different intervals for each slave node to avoid request bursts. We adopted this method to optimize the parallel Stochastic Gradient Descent algorithm, through which we successfully sped up parameter-exchanging process by eight times. [ABSTRACT FROM AUTHOR]

The new emergence of data-intensive applications raises a huge requirement on the capacity of memory. However, an obvious increase of memory makes the corresponding energy consumption unacceptable in practice. To address this question, any effort only to reduce the energy consumption of dynamic random access memory (DRAM) is ineffective. Recently, combining DRAM and phase change memory (PCM) to construct a hybrid main memory is recognized as a promising solution to distinctly reduce the energy consumption. In this paper, we propose a new page-level energy management strategy to optimize the energy consumption of the hybrid main memory. The proposed strategy records pages' local and global access information by a new data structure, and then classifies pages by the access history, at last adaptively places PCM or DRAM pages according to the memory characteristics and remaps the migrated pages. Our experimental results show that our strategy can achieve 9.4 % of energy saving and 9.6 % of performance improvement at most compared with APG and PDRAM, which were proposed respectively by conferences RACS'12 and DAC'09. [ABSTRACT FROM AUTHOR]

In multi-program environment, cache contention among processors can significantly degrade system performance. Cache partitioning served as an effective measure has been widely studied, especially for dynamic cache partitioning. However, it is difficult to decide the best cache quota which should be allocated to co-scheduled programs and the best time when a cache adjusting should be performed in dynamic cache partitioning scheme. This paper presents a novel dynamic cache partitioning mechanism based on the phase behavior of programs. It uses the performance monitoring units of modern processors and detects the phase behavior of programs to guide the cache partitioning at run-time. Since programs have recurring phase behavior during the whole execution time, on one hand, we can adjust the cache quota when a phase change occurs, on the other hand, we can make cache partitioning policy with higher accuracy and lower overhead by classifying phases. The method proposed in this work is validated in the measured results for applications from SPEC CPU 2006 benchmark suite. Compared with the performance of shared cache scheme, our method can achieve a speedup up to 1.214 for co-scheduled applications. [ABSTRACT FROM AUTHOR]

Edge computing is highly recommended to support Mobile Crowdsensing (MCS) applications for sensing data processing. In this paper, we consider the MCS applications supported by the mobile phones of bus passengers, who transfer between different bus stations equipped with edge servers. The edge servers deployed with the corresponding MCS services can acquire and process the sensing data directly from the participants' sensors by device-to-device (D2D) communication. Therefore, with the help of edge services, it is desirable to deploy more MCS services to explore more D2D communications, without incurring cellular communication cost. Different stations are with different the number of passengers passing through, resulting in different benefits of edge service deployment. Taking the bus passenger mobility characteristics into consideration, we shall seek a tradeoff between the communication cost and service deployment cost to pursue overall cost efficiency. We first formulate the problem into a mixed integer linear programming model and then design a low-complexity heuristic algorithm. Performance evaluation results verify the high efficiency of our algorithm by the fact that it can much approach the optimal solution. [ABSTRACT FROM AUTHOR]

It is meaningful to use a little energy to obtain more performance improvement compared with the increased energy. It also makes sense to relax a small quantity of performance restriction to save an enormous amount of energy. Trading a small amount of energy for a considerable sum of performance or vice versa is possible if the relativities between performance and energy of parallel programs are exactly known. This work studies the relativities by recording the performance speedup and energy consumption of parallel programs when the number of cores on which programs run are changed. We demonstrate that the performance improvement and the increased energy consumption have a linear negative correlation.In addition, these relativities can guide us to do performance-energy adaptation under two assumptions. Our experiments show that the average correlation coefficients between performance and energy are higher than 97 %. Furthermore, it can be found that exchanging less than 6 % performance loss for more than 37 % energy consumption is feasible and vise versa. [ABSTRACT FROM AUTHOR]

More and more unstructured data are produced and consumed over network. How to maintain these data and improve the availability and scalability of the storage systems has become a considerable challenge. Although some NoSQL systems such as Dynamo, Cassandra, MongoDB have provided different advantages for unstructured data management, no one can provide flexible query functions like MongoDB, meanwhile guarantee the availability and scalability as Cassandra simultaneously. This paper presents a new high available distributed storage system called MyStore based on an optimized clustered MongoDB for unstructured data. Consistent hash is used to distribute data on multiple MongoDB nodes by applying virtual node method. NWR mode is applied to provide automatic backup operation and guarantee data consistency. And a gossip protocol is taken for exchanging information of failures in the system. Moreover, a user-friendly interface module and an efficient cache module are designed for improving the usability of the system. Based on above strategies, the system can realize high availability for unstructured data storage, while providing complex query functions like rational databases. Moreover, it is applied in a multi-discipline virtual experiment platform named VeePalms that has run practically. Experimental evaluation shows that the methodology is powerful enough not only to enhance the data availability, but also to improve the server's scalability. [ABSTRACT FROM AUTHOR]

Recently, BitTorrent communities are rapidly evolving into private torrent sites (PT). PT sites have employed several incentive rules to improve the performance and availability of the system. Many studies have been dedicated for measuring and modeling the PT systems in order to better understand the new rules and their impact on the users' behavior in order to improve the usability of the system. These studies have been performed on different PT sites that differ in their implementation of the system and in their system's user incentive rules. Therefore, current measurement findings cannot reflect accurate results and, more importantly, the current conclusions may be biased. In this paper, we investigate the accuracy of previous measurement studies on PT sites, while emphasizing the incentive rules employed and the interplay between these rules and corresponding objective factors. We evaluate the behavior regulation policies of the front-end website and the tracker and examine the semantics of provided data. Using this information we have designed a new crawling methodology and conducted a large-scale measurement study across four representative PT sites over a year. Interestingly, we find that most reported measurements have neither considered design goals, nor thought through the incentive policies and their interplay. This lack of awareness in turn may lead to inaccurate conclusions of system properties. For example, the Seeder to Leecher Ratio (SLR), which is reported in most of the available measurements, is routinely at least 16-45 % less than the real SLR because the sites ignored the 'partial seeders' in their calculation. This study aims to offer fundamental insights into designing an accurate methodology when conducting measurement studies on PT sites. [ABSTRACT FROM AUTHOR]

Thanks to the availability of various mobile applications, lots of users shift from desktop environments, e.g., PCs and laptops, to mobile devices, e.g., smartphones and tablets. However, there are still some desktop applications without counterparts on mobile devices, such as some integrated development environments (e.g., eclipse) and automatic industry control systems. In this paper, we propose Modeap, a platform-independent mobile cloud service that can push all desktop applications developed for various operating systems from cloud servers to mobile devices. Modeap follows a design principle of complete detachment and regeneration of desktop user interface, i.e., the essential graphical primitives of the original desktop applications will be intercepted and then translated into standard web-based graphical primitives such that the interactions between users and remote cloud applications become possible via mobile web browsers. In this way, all desktop applications built upon the same set of graphical primitives can be used on mobile devices in great flexibility without installing any new software. We have developed a proof-of-concept prototype that provides Windows applications from cloud server to mobile web browsers. The results of extensive experiments show that the proposed framework can achieve our design goals with low latency and bandwidth consumption. [ABSTRACT FROM AUTHOR]

Live migration of virtual machine (VM) provides a significant benefit for virtual server mobility without disrupting service. It is widely used for system management in virtualized data centers. However, migration costs may vary significantly for different workloads due to the variety of VM configurations and workload characteristics. To take into account the migration overhead in migration decision-making, we investigate design methodologies to quantitatively predict the migration performance and energy consumption. We thoroughly analyze the key parameters that affect the migration cost from theory to practice. We construct application-oblivious models for the cost prediction by using learned knowledge about the workloads at the hypervisor (also called VMM) level. This should be the first kind of work to estimate VM live migration cost in terms of both performance and energy in a quantitative approach. We evaluate the models using five representative workloads on a Xen virtualized environment. Experimental results show that the refined model yields higher than 90% prediction accuracy in comparison with measured cost. Model-guided decisions can significantly reduce the migration cost by more than 72.9% at an energy saving of 73.6%. [ABSTRACT FROM AUTHOR]