Recent years have witnessed an exponential growth of model scale in deep learning-based recommender systems---from Google's 2016 model with 1 billion parameters to the latest Facebook's model with 12 trillion parameters. Significant quality boost has come with each jump of the model capacity, which makes us believe the era of 100 trillion parameters is around the corner. However, the training of such models is challenging even within industrial scale data centers. We resolve this challenge by careful co-design of both optimization algorithm and distributed system architecture. Specifically, to ensure both the training efficiency and the training accuracy, we design a novel hybrid training algorithm, where the embedding layer and the dense neural network are handled by different synchronization mechanisms; then we build a system called Persia (short for parallel recommendation training system with hybrid acceleration) to support this hybrid training algorithm. Both theoretical demonstrations and empirical studies with up to 100 trillion parameters have been conducted to justify the system design and implementation of Persia. We make Persia publicly available (at github.com/PersiaML/Persia) so that anyone can easily train a recommender model at the scale of 100 trillion parameters.

Energy efficiency of data centers has draw a great attention due to the cost of power consumption increases dramatically as the size of data center grows. Nowadays, Map Reduce is a framework widely used for processing large data sets in data center, its energy efficiency directly affects the energy efficiency of data center. MapReduce's energy efficiency is closely tied to its scheduler, we find that fair scheduler outperforms FIFO scheduler in energy efficiency when CPU-intensive job and IO-intensive job running simultaneously on the cluster, because fair scheduler achieves better resource utilization by overlapping resource complementary tasks on slaves. However this behavior is occasional, because fair scheduler has no information about task's resource requirement. This occasional behavior lets us identify the area that energy efficiency of fair scheduler can be improved. We propose an energy-efficient scheduling policy called green scheduling which relaxes fairness slightly to create as many opportunities as possible for overlapping resource complementary tasks. The results show that green scheduling can save between 7% and 9% energy consumption of fair scheduler.

Large datasets become common in applications like Internet services, genomic sequence analysis and astronomical telescope. The demanding requirements of memory and computation power force data mining algorithms to be parallelized in order to efficiently deal with the large datasets. This paper introduces our experience of grouping internet users by mining a huge volume of Web access log of up to 100 gigabytes. The application is realized using hierarchical clustering algorithms with Map-Reduce, a parallel processing framework over clusters. However, the immediate implementation of the algorithms suffers from efficiency problem for both inadequate memory and higher execution time. This paper present an efficient hierarchical clustering method of mining large datasets with Map-Reduce. The method includes two optimization techniques: ¿Batch Updating¿ to reduce the computational time and communication costs among cluster nodes, and ¿Co-occurrence based feature selection¿ to decrease the dimension of feature vectors and eliminate noise features. The empirical study shows the first technique can significantly reduce the IO and distributed communication overhead, reducing the total execution time to nearly 1/15. Experimentally, the second technique efficiently simplifies the features while obtains improved accuracy of hierarchical clustering.

In computational grids, a virtual organization (VO) is a dynamic coupling of multiple Linux/Unix nodes for resource sharing under specific polices. Currently, VO support functionalities are generally implemented as grid middleware. However, the usability of grids is often impaired by the complexity of configuring and maintaining a new layer of security infrastructure as well as adapting to new interfaces of security enabled services. In this paper, we present an OS-level approach to provide native VO support functionalities, which is a part of XtreemOS project [18]. Our approach adopts pluggable frameworks existing in current OS as extension points to implement VO support, avoiding modification of kernel codes and easily turning traditional OSes into grid-aware ones. The performance evaluation of NAS parallel benchmarks (NPB) shows that our current implementation incurs trivial overhead on original systems.

Grid and service computing technologies have been explored by enterprises to promote integration, sharing, and collaboration. However, quick response to business environment changes is still a challenging issue. For end users, developing, customizing, and reengineering applications remain a difficult and timeconsuming task. Users still need to deal with excessive low-level details of platform-specific APIs. We present a high-level programming model together with a descriptive glueing language called GSML, to facilitate end-user programming. In this approach, applications could be visually composed from welldefined software components called "funnels" in an event-driven fashion. Application examples have shown that, by raising the level of abstraction as well as simplifying the programming model, GSML could empower end users to build grid applications on demand with improved productivity.