We have been developing a multiprocessor architecture which executes iterations of loop speculatively in parallel. In this paper, we present speculative memory (SM), order to enable the large-scale speculation supports execution iteration arbitrary size and duration. With SM, programmer can hint explicitly that certain are preferable be executed SM manages multiple values (versions) modified data. also features renaming delayed program codes, could viewed as dynamic code migration. These...

In this paper, we present an e-learning back-end system which cooperates with a learning management (LMS). Our is aimed at fair and effective assessment in class of elementary programming practice. While most LMSs provide basic functions to support various courses, complements such conventional by providing the following specialized functions: syntax check codes, plagiarism detection, automated black-box testing. makes general-purpose LMS more useful for both students teachers low cost.

We have been developing a multiprocessor architecture which creates speculative threads from sequential program and executes them in parallel. In this architecture, we aim at the large-scale speculation supports execution of arbitrary size duration. So, our system must be able to analyze dependency on large amounts memory data. paper, describe outline current design mechanism for dynamic inter-thread analysis, renaming, data management detail. These mechanisms not only enables amount...

Many techniques for parallelizing a sequentially coded program have been developed and put to practical use, but there are many cases in which codes cannot be parallelized because it is impossible assure that their parallel execution does not violate the data dependencies program. To parallelize such programs, we previously proposed speculative memory (SM). With SM, programmers can specify of threads explicitly programs. The SM system manages speculatively read or written by running...

We had proposed a distributed storage system which relocates data blocks autonomously among the nodes. Generally, owing to requirement of exchanging management information frequently, such autonomous block migration tends increase network traffic. Our previous work has presented scheme reduce traffic by appending onto an iSCSI short packet. In this paper, we present propagate utilizing internode communications more aggressively. utilizes read/write packets are exchanged frequently transmit...

In our previous work, we had proposed a distributed storage system which aims to optimize the locations of data blocks on basis their access frequency. order relocate at an appropriate node, each node should collect fresh information about utilization other nodes. However, frequent exchange might become cause network congestion. this paper, propose method improve freshness and suppress traffic. This controls propagation interval according condition node. We also show that Ethernet frames can...

Thread-Level Speculation (TLS) is an approach to enhance the opportunity of parallelization by executing tasks in parallel based on assumption that task has no dependencies any earlier program order. But if dependency detected during execution, should be aborted and re-executed. So frequency aborts one factors damage performance speculative execution. In this paper we propose "code shelving" scheme avoid or eliminate penalty abort. We have implemented it our TLS system, which named...

We propose a distributed storage system which relocates data blocks autonomously among the nodes.In order to optimize I/O performance with slight administrative workload, our is aimed at realization of following two functions; (i) Run-time construction tiers from heterogeneous devices, and (ii) Automated block migration tiers.We also show technique reduce management traffic in iSCSI environment.

We have proposed a distributed storage system which dynamically makes tiers and optimizes location of data blocks autonomously. This aims to enhance the I/O performance without remarkable network overhead. Our organizes considering device characteristics. And will be placed in suitable tier according their access pattern.In this paper, we propose method select destination node for migration using an characteristic block migrated. ranks nodes, autonomously configures based on relative...

In this paper, we propose a distributed storage system which relocates data blocks autonomously among the nodes. is composed of heterogeneous nodes, there are performance gaps devices. The tiers made by such in system. When frequently accessed placed on improper tier, may decline. To avoid problem, administrator should periodically investigate state utilization and relocate to proper node. Although makes efficient use devices, administrators' workload quite heavy. Our aimed at realization...

In this paper, we propose a new Hardware Transactional Memory (HTM) system for shared-memory multiprocessor in which elementary processors are connected by single common bus. One of the key features our is modified snoop cache protocol to reduce overheads on transactional memory consistency control. By publishing all data transaction at once when commits, avoids overhead commit, would arise from sequential publication (or write-back main memory) each item otherwise. Another feature...

Thread-level speculation (TLS) is an approach to enhance the opportunity of parallelization programs. A TLS system enables multiple threads begin execution tasks in parallel even if there may be dependency between tasks. When any violation detected, enforces violating thread abort and re-execute task. So, frequency aborts one factors that damage performance speculative execution. This article proposes a new technique named code shelving, which not need abort. It available only for but also...

Transactional Memory (TM) is promising to make parallel programming easier. There have been many hardware implementations of transactional memory (HTM) proposed improve the performance, but they still suffer from some overheads when a transaction commits or aborts. So, we developing novel new HTM design, called DCTM, which enables transactions arbitrary size commit abort in fixed number cycles -- typically one cycle. DCTM stores values data modified into an L1 cache. Each cache checks...

To enlarge the opportunities for parallelizing a sequentially coded program, authors have previously proposed speculative memory (SM). With SM, they can start parallel execution of program by assuming that it does not violate data dependencies in program. When SM system detects violation, recovers computational state and restarts execution. In this article, explore design space implementing software-based system. They compared possible choices following three viewpoints: (1) which waiting...

k-dimensional (k-d) trees are one of the most important data structures in fields engineering and so-called Big Data. In this paper we propose a scheme parallelizing construction k-d tree. Since efficient presorting is required for constructing balanced tree, also developed parallelized heapsort algorithm. The proposed 3.59 times faster than sequential

Heapsort algorithm was found many decades ago, but no parallel heapsort had been presented. In our previous work, we challenged the development of a algorithm. this paper improve partially and analyze performance in detail. As more massive amount data are sorted, or as complex comparison is required for sorting, can sort efficiently.

We have been developing a software-based Speculative Memory (SM) library to enlarge the opportunity for program parallelization. It enables tasks having possible unknown dependencies on each other be executed speculatively in parallel while dynamically inspecting among those tasks. When violation of (misspeculation) is detected, it aborts execution violating task and re-executes it. But re-execution from its beginning may bring large performance losses. To reduce such losses, therefore, this...

Our previous works showed a distributed storage system capable to migrate data blocks autonomously across tiers. In this paper, we present an adaptive method for constructing tiers in the autonomous system. This can reduce task tuning runtime parameters. The normally depends on configuration of devices and behavior process, so it should be done periodically by administrators. proposed fine-tunes during execution match environment, thus eliminating need manual parameter adjustments...

We have proposed a distributed storage system which dynamically makes tiers and optimizes location of data blocks autonomously. This aims to enhance the I/O performance without remarkable network overhead. Our organizes considering device characteristics. And will be placed in suitable tier according their access pattern. In this paper, we propose method select destination node for migration using an characteristic block migrated. ranks nodes dynamically, each configures Simulation results...

Binary search trees (BSTs) are one of the most important data structures in computer science. A parallel construction algorithm a BST can be easily derived from sequential algorithm. Since structure generated is different depending on order inserted nodes, however, such cannot generate with same (node position) as by So this paper presents new to construct having constructs. This was based concept thread-level speculation but purely (non-speculatively) one. Our experiments showed that for...

Sorting is a fundamental and essential problem required in the wide range of application fields, so many sorting algorithms have been developed. Among those algorithms, heapsort one most elegant efficient algorithms. But no parallel algorithm had presented until authors developed restricted few years ago. This restriction which makes it difficult to be used universally for general data sets. So, this article, present modified free from such can any set. new achieve almost same performance as before.