Search

Programs now have more aggressive demands of memory to hold their data than before. This paper analyzes the characteristics of memory data by using seven real memory traces. It observes that there are a large volume of memory pages with identical contents contained in the traces. Furthermore, the unique memory content accessed are much less than the unique memory address accessed. This is incurred by the traditional address-based cache replacement algorithms that replace memory pages by checking the addresses rather than the contents of those pages, thus resulting in many identical memory contents with different addresses stored in the memory. For example, in the same file system, opening two identical files stored in different directories, or opening two similar files that share a certain amount of contents in the same directory, will result in identical data blocks stored in the cache due to the traditional address-based cache replacement algorithms. Based on the observations, this paper evaluates memory compression and memory deduplication. As expected, memory deduplication greatly outperforms memory compression. For example, the best deduplication ratio is 4.6 times higher than the best compression ratio. The deduplication time and restore time are 121 times and 427 times faster than the compression time and decompression time, respectively. The experimental results in this paper should be able to offer useful insights for designing systems that require abundant memory to improve the system performance.

Programs now have more aggressive demands of memory to hold their data than before. This paper analyzes the characteristics of memory data by using seven real memory traces. It observes that there are a large volume of memory pages with identical contents contained in the traces. Furthermore, the unique memory content accessed are much less than the unique memory address accessed. This is incurred by the traditional address-based cache replacement algorithms that replace memory pages by checking the addresses rather than the contents of those pages, thus resulting in many identical memory contents with different addresses stored in the memory. For example, in the same file system, opening two identical files stored in different directories, or opening two similar files that share a certain amount of contents in the same directory, will result in identical data blocks stored in the cache due to the traditional address-based cache replacement algorithms. Based on the observations, this paper evaluates memory compression and memory deduplication. As expected, memory deduplication greatly outperforms memory compression. For example, the best deduplication ratio is 4.6 times higher than the best compression ratio. The deduplication time and restore time are 121 times and 427 times faster than the compression time and decompression time, respectively. The experimental results in this paper should be able to offer useful insights for designing systems that require abundant memory to improve the system performance.

“Variety” in Big Data means we have a wide range of data types and sources: e.g. file systems and database systems co-exist for decades as two popular data-accessing interfaces. This work is to unify these two interfaces by presenting a Data Interface All-iN-A-place (DIANA). The first challenge lies in distinguishing structured and un-structured data and diverting them to different underlying platforms. It is demonstrated that a speedup of 5000 in indexing has been achieved at the expense of a slowdown of 100 in extracting attributes. A DIANA-based cloud storage system is constructed for versatile, long distance and large volume big data accessing operations to address “Volume” and “Velocity” in Big Data. It encapsulates a dynamic multi-stream/multi-path engine at the socket level, which conforms to Portable Operating System Interface (POSIX).

This paper presents a framework of Mirroring Mobile Phone in the Clouds (MMPC) to speed up data/computing intensive applications on a mobile phone by taking full advantage of the super computing power of the clouds. An application on the mobile phone is dynamically partitioned in such a way that the heavy-weighted part is always running on a mirrored server in the clouds while the light-weighted part remains on the mobile phone. A performance improvement (an energy consumption reduction of 70% and a speed-up of 15x) is achieved at the cost of the communication overhead between the mobile phone and the clouds (to transfer the application codes and intermediate results) of a desired application. Our original contributions include a dynamic profiler and a dynamic partitioning algorithm compared with traditional approaches of either statically partitioning a mobile application or modifying a mobile application to support the required partitioning.

Spatial On-LineAnalyticalProcessing(SOLAP)isapowerfuldecisionsupportsystemstoolforexploring\ud the multidimensionalperspectiveofspatialdata.Inrecentyears,remotelysenseddatahavebeen\ud integratedintoSOLAPcubes,andthisimprovementhasadvantagesinspatio-temporalanalysisfor\ud environmentmonitoring.However,theperformanceofaggregationsinSOLAPstillfacesaconsiderable\ud challenge fromthelarge-scaledatasetgeneratedbyEarthobservation.Fromtheperspectiveofdata\ud parallelism, atile-basedSOLAPcubemodel,theso-calledTileCube,ispresentedinthispaper.Thenovel\ud model implementsRoll-Up/Drill-AcrossoperationsintheSOLAPenvironmentbasedonMap-Reduce,a\ud popular data-intensivecomputingparadigm,andimprovesthethroughputandscalabilityofraster\ud aggregation. Therefore,thelongtime-series,wide-rangeandmulti-viewanalysisofremotelysensed\ud data canbeprocessedinashorttime.TheTileCubeprototypewasbuiltonHadoop/Hbase,anddrought\ud monitoring isusedasanexampletoillustratetheaggregationsinthemodel.Theperformancetesting\ud indicated themodelcanbescaledalongwithboththedatagrowthandnodegrowth.Itisapplicableand\ud natural tointegratetheSOLAPcubewithMap-Reduce.Factorsthatinfluence theperformancearealso\ud discussed, andthebalanceofthemwillbeconsideredinfutureworkstomakefulluseofdatalocalityfor\ud model optimisation.

Memristor is short for memory resistor, which provides a functional relation between flux and charge. Professor Leon Chua named and formulated it in his paper\ud "Memristor—The Missing Circuit Element" in 1971. The memristor has a special effect, ‘the delayed switching effect’, which is the memristor switching takes place with a time delay. Content addressable memory (CAM) is a type of associative memories that is adopted in high speed searching applications. The new Memristor-CAM cell has been proposed, which included two memristors; one is used as an important part of the comparator that is instead of the traditional logic gate; and another one is for storing the data as a storage element. In this paper, we report that applying the memristor delayed switching effect to the novel design of the Memristor Content Addressable Memory (M-CAM) cell. The delayed switching effect is used to control the changing time of the memristor’s state, which can enhance the performance, decrease the searching time and save energy.