To design next-generation Graphics Processing Units (GPUs), GPU architects rely on GPU performance analyses to identify key GPU performance bottlenecks and explore GPU design spaces.Unfortunately, the existing GPU performance analysis mechanisms make it difficult for GPU architects to conduct fast and accurate GPU performance analyses.The existing mechanisms can provide misleading

Processing-In-Memory (PIM) is an attractive solution for mitigating frequent and large data movement between computational units and memory devices. Among various PIM implementations, Samsung Aquabolt-XL is an HBM2 memory device which implements 16 PIM-enabled pseudo-channels and associates an In-Memory Processor (IMP) to each pair of the memory banks. Recent studies have shown that Aquabolt-XL can greatly accelerate various applications (e.g., deep learning) by offloading memory-intensive operations (e.g., matrix-vector multiplications) to the IMPs. However, the prior study fails to fully utilize Aquabolt-XL and achieves limited performance gains by offloading operations to the IMPs of only a single pseudo-channel. Ideally, utilizing all the 16 pseudo-channels of Aquabolt-XL can further accelerate the key operations by a factor of 16× compared to utilizing only a single pseudo-channel. To fully exploit Aquabolt-XL, therefore, memory-intensive operations should be offloaded to and concurrently executed on the IMPs of all the PIM-enabled pseudo-channels. This paper presents MPC-Wrapper, a multi-pseudo-channel wrapper interface which allows memory-intensive operations to be offloaded to and concurrently executed on the IMPs of all the 16 PIM-enabled pseudo-channels of Aquabolt-XL. First, MPC-Wrapper allows all the PIM-enabled pseudo-channels to operate independently and in parallel, thus achieving high scalability needed for fully utilizing all the PIM-enabled pseudo-channels of Aquabolt-XL. Second, MPC-Wrapper is highly flexible as it exposes the PIM-enabled pseudo-channels as separate ports and enables an FPGA logic to flexibly utilize any set of the PIM-enabled pseudo-channels according to its needs. Third, MPC-Wrapper achieves high usability by hiding the complex low-level interactions between the memory controller and Aquabolt-XL for initializing and invoking the PIM-enabled pseudo-channels from the other FPGA logics. Using an Aquabolt-XL-equipped Xilinx Alveo U280 FPGA and four memory-intensive benchmarks, we show that utilizing all the 16 PIM-enabled pseudo-channels of Aquabolt-XL with MPC-Wrapper achieves a geometric mean speedup of 13.66× over the baseline single PIM-enabled pseudo-channel implementations of the benchmarks.

Apache Hadoop is a famous framework for big data science. Most of the research for improving the speed of big data analysis is researching based on Hadoop modules such as Hadoop common, Hadoop Distribute File System (HDFS), Hadoop Yet Another Resource Negotiator (YARN) and Hadoop MapReduce. The paper focuses on data locality on HDFS and MapReduce to improve the performance. The input data is divided into several blocks and stored in HDFS. Each block has sever key-value fair in map stages. The paper use the keys in block to make key-based Deep Data Locality (DDL). The MapReduce with key-based DDL reduce some steps on map stage, shuffle stage, and reducer stages to improve the performance of MapReduce. We tested the performance of MapReduce with block-based DDL and key-based DDL to compare with default MapReduce. According to the test, MapReduce with key-based DDL is 28% faster than default MapReduce and 15.4 % faster than MapReduce with block-based DDL. Additionally, key-based DDL can be combined other data locality methods to improve the Hadoop. Combined key-based DDL and block-based DDL improve the Hadoop performance up to 52.5%.The paper also introduced the simulator for testing the performance of MapReduce with applied data locality methods on Hadoop. The simulator display a performance of each stage of MapReduce using graph. Key-based DDL can be combined with other data locality research to get optimized performance in various data types and node' status in the simulator.

MapReduce has been widely used in many data science applications. It has been observed that an excessive data transfer has a negative impact on its performance. To reduce the amount of data transfer, MapReduce utilizes data locality. However, even though the majority of the processing cost occurs in the later stages, data locality has been utilized only in the early stages, which we call Shallow Data Locality (SDL). As a result, the benefit of data locality has not been fully realized. We have explored a new concept called Deep Data Locality (DDL) where the data is pre-arranged to maximize the locality in the later stages. Toward achieving stronger DDL, we introduce a new block placement paradigm called Limited Node Block Placement Policy (LNBPP). Under the conventional default block placement policy (DBPP), data blocks are randomly placed on any available slave nodes, requiring a copy of RLM (Rack-Local Map) blocks. On the other hand, LNBPP places the blocks in a way to avoid RLMs, reducing the block copying time. The containers without RLM have a more consistent execution time, and when assigned to individual cores on a multicore node, they finish a job faster collectively than the containers under DBPP. LNBPP also rearranges the blocks into a smaller number of nodes (hence Limited Node) and reduces the data transfer time between nodes. These strategies bring a significant performance improvement in Map and Shuffle. Our test result shows that the execution times of Map and Shuffle have been improved by up to 33% and 44% respectively. In this paper, we describe the MapReduce workflow in Hadoop with a simple computational model and introduce the current research directions in each step. We analyze the block placement status and RLM locations in DBPP with the customer review data from TripAdvisor and measure the performances by executing the Terasort Benchmark with various sizes of data. We then compare the performances of LNBPP with DBPP.

Recently, majority of environmental monitoring systems use Internet of Things devices [1]. The Nevada Solar Energy-Water-Environment Nexus project collects a large amount of environmental data from various remote sensors. In the previous research, we introduced REST API for communication and web service between senor and server [2]. However, current RESTful web service has several issues for authentication system. We developed an ID-based authentication algorithm to achieve a secure RESTful web service. ID-based authentication is based on Boneh-Franklin ID-based encryption and REST URI. Such process enabled server to handle client's request by acknowledging client's URI rather than storing client's entire status for stateless REST. Also, such process resolves current problems generated by implementing REST authentication.

Authentication to a computing system is composed of two parts, identification and verification. Traditionally, login IDs have been used for identification and passwords for verification. Many schemes have been proposed to improve both parts, but they may require specialized devices or they may not be always reliable. We propose a method that can augment the current password-based system by strengthening the identification process. It utilizes personal secret data instead of a login ID to identify a user uniquely, hence mindmetrics. It then asks the user to choose a correct login ID among multiple choices of partially obscured IDs. Since it does not accept a login ID during the authentication process, a stolen or cracked password cannot be used for gaining an access to the computing system unless the attacker provides a correct identification material, i.e., mindmetrics token. This additional step raises the security of an authentication system considerably over single or double password systems. Since the stolen passwords cannot be used immediately by the attackers, account holders can have extra time to change their passwords before the attackers gain an access. This scheme does not require any specialized hardware and can be implemented easily. It may be used where biometrics schemes cannot be used cost-effectively, e.g., on public e-commerce web sites. Mindmetrics scheme separates the identification server and the verification server, thus it is scalable to a large system. We implemented a proof-of-concept system and evaluated it with test users. The survey indicates that the system is not as intrusive as other schemes, users feel better protected, and they are willing to use the scheme on public web sites.

Many scientific applications generate massive data that requires visualization. For example, the Nevada Solar Energy-Water-Environmental Nexus project has been generating a large amount of environmental monitoring data in textual format. As the data is available on the web, a web-based visualization tool is desirable for the project rather than a standalone tool. This research analyzes the processing mechanisms of four popular web-based data visualization tools, that is, Google Charts, Flex, OFC, D3, and compares their performances. A standalone visualization tool, JfreeChart, have been also used for comparison. The processing times have been divided into three segments, layout time, data transformation time, and rendering time, and separately measured. The actual temperature data from the Nevada Nexus project has been used for testing in different scales ranging from 100 to 100,000 data points. The result shows that each visualization tool has its own ideal environment.

Silicon validation becomes difficult because of rapidly increasing complexity and operation speed of integrated circuits. When an error is found after a chip is fabricated, post-silicon repair is necessary. Full mask revision may significantly increase the cost and time-to-market. In this paper, we describe partial metal revision techniques in which only top-level metal layers are revised to fix ¿small¿ errors with minimal increase of the cost. When an error cannot be fixed by partial metal layer revision, full metal revision or full mask revision is necessary. However, frequently errors are small enough to be fixed by partial metal layer revision. Effective partitioning and pin-extension to top-level metal layers can significantly improve the repairability by using top-level metal revision.