The long-term use of machine learning models can result in degraded performance due to data drift and other factors. We have previously proposed a data selection mechanism for time-series data of machine learning models. When data drift occurs, the models have to learn again using large-scale stream data. Thus, it is important for machine learning algorithms to introduce a mechanism avoiding useless data. This study examines the effect of data selection with an adversarial classifier using synthetic data.

Rapidly escalating energy and cooling costs of storage systems have become a concern for data centers. In response, a multitude of energy saving approaches that take into account storage-device-level input/output (I/O) behaviors has been proposed. The trouble is that critical applications are in constant operation at data centers, and the conventional approaches do not produce sufficient energy savings. It may be possible to dramatically reduce storage energy consumption without degrading application performance levels by utilizing application level I/O behaviors. However, such behaviors differ from one application to another, and it would be too expensive to tailor methods to individual applications. We propose a universal storage energy management framework for runtime storage energy savings that can be applied to any type of application. The results of evaluations show that the use of this framework results in substantive energy savings compared with the traditional approaches.

Rapidly escalating energy and cooling costs, especially those related to the energy consumption of storage systems, have become a concern for data centers, primarily because the amount of digital data that needs storage is increasing daily. In response, a multitude of energy saving approaches that take into account storage-device-level input/output (I/O) behaviors have been proposed. The trouble is that numerous critical applications such as database systems or web commerce applications are in constant operation at data centers, and the conventional approaches that only utilize storage-device-level I/O behaviors do not produce sufficient energy savings. It may be possible to dramatically reduce storage-related energy consumption without degrading application performance levels by utilizing application-level I/O behaviors. However, such behaviors differ from one application to another, and it would be too expensive to tailor methods to individual applications. As a way of solving this problem, we propose a universal storage energy management framework for runtime storage energy savings that can be applied to any type of application. The results of evaluations show that the use of this framework results in substantive energy savings compared with the traditional approaches that are used while applications are running.

Power, especially that consumed for storing data, and cooling costs for data centers have increased rapidly. The main applications running at data centers are data intensive applications such as large file servers or database systems. Recently, power management of the data intensive applications has been emphasized in the literature. Such reports discuss the importance of power savings. However, these reports lack research on power management models for the efficient use of data intensive applications' I/O behaviors. This paper proposes a novel energy efficient storage management system that monitors both application- and device-level I/O patterns at run time, and uses not only the device-level I/O pattern but also application level patterns. First, the design of the proposed model combined with such large data intensive applications will be shown. The key features of the model are i) classifying application-level I/O into four patterns using run-time access behaviors such as the length of idle time and read/write frequency, and ii) adopting an appropriate power-saving method-based on these application level I/O patterns. Next, the proposed method is quantitatively evaluated with typical data intensive applications such as file servers, OLTP, and DSS. It is shown that energy efficient storage management is effective in achieving large power savings compared with traditional approaches while an application is running.

The distributed shared memory (DSM) architecture is considered to be one of the most likely parallel computing environment candidate for the near future because of its ease of system scalability and facilitation for parallel programming. However, a naive program based on shared memory execution on a DSM machine often deteriorates performance, because of the overhead involved for maintaining cache coherency particularly with frequent remote memory accesses. We show that careful buffer management of parallel join processing on DSM can produce considerable performance improvements in comparison with a naive implementation. We propose four buffer management strategies for parallel hash join processing on the DSM architecture and actually implement them on the HP Exemplar SPP 1600. The basic strategy is to begin with the hash join algorithm for the shared everything architecture and then to consider the memory locality of DSM by distributing the hash table and data pool buffers among the nodes. The results of four buffering strategies are analyzed in detail. Consequently, we can conclude that, in order to achieve high performance on a DSM machine, our buffer management strategy in which the memory access pattern is extracted and buffers are allocated in the local memory of nodes to minimize memory access cost is very efficient.

The authors implemented a parallel hash join algorithm on a Symmetry S81 shared-everything multiprocessor environment and evaluated the performance. They evaluated the input/output (I/O) performance on a multiple-disk environment, and showed linear performance increase of up to eight disks. The performance of the implemented join operation was examined on each phase, and the effect of parallel processing by the multiprocessor and the multiple disks was clarified. It was concluded from the experimental result that on such a shared-everything multiprocessor system parallelism could be easily exploited for the construction of high-performance relational database systems.>

The performance of the functional disk system with relational database engine (FDS-R2) is evaluated in detail in view of two points. First, the performance evaluation of the combined hash algorithm on FDS-R2 is reported using the projection and aggregation operations in addition to the join operation and they are analyzed in order to verify the effectiveness of the proposed processing method, Second, several measured results of performance evaluations with the expanded version of the Wisconsin Benchmark are given and analyzed. FDS-R2 attained higher performance for very large relations as compared to other large database systems such as Gamma and Teradata. In this evaluation, it is also shown that the performance of the relational operations can be improved largely by using an efficient hashing strategy for large relations on FDS-R2.>

The second version of FDS-R (functional disk system with relational database engine), FDS-RII, which is designed to handle large relations efficiently, is discussed. On FDS-RII, the processing algorithm is selected at run time from two algorithms (nested loop algorithms, grace hash algorithm) by comparing their estimated I/O costs. The processing strategy is discussed in detail. The I/O cost formula is examined by measuring the execution time of a join query on the FDS-RII. With the expanded version of Wisconsin Benchmark, the performance of FDS-RII is measured. FDS-RII attained a high performance level for large relations as compared to other large database systems such as Gamma and Teradata. While FDS uses just one disk and three MC68020s, Teradata uses 40 disks and 20 AMPs and Gamma requires eight disks and 17 VAX 11/750s.>

The major performance bottle neck in the current computer system is in the low-performance secondary system. The performance of the CPU has increased dramatically so far, about several orders of magnitude improvement has been achieved. On the other hand, that of the disk system has shown little advance since nineteen sixties. The von Neumann bottle neck between the CPU and the secondary storage subsystem has been much more enlarged.