scholarly journals Pengukuran Performa Apache Spark dengan Library H2O Menggunakan Benchmark Hibench Berbasis Cloud Computing

2019 ◽  
Vol 6 (5) ◽  
pp. 519
Author(s):  
Aminudin Aminudin ◽  
Eko Budi Cahyono

<p class="Judul2">Apache Spark merupakan platform yang dapat digunakan untuk memproses data dengan ukuran data yang relatif  besar (<em>big data</em>) dengan kemampuan untuk membagi data tersebut ke masing-masing cluster yang telah ditentukan konsep ini disebut dengan parallel komputing. Apache Spark mempunyai kelebihan dibandingkan dengan framework lain yang serupa misalnya Apache Hadoop dll, di mana Apache Spark mampu memproses data secara streaming artinya data yang masuk ke dalam lingkungan Apache Spark dapat langsung diproses tanpa menunggu data lain terkumpul. Agar di dalam Apache Spark mampu melakukan proses machine learning, maka di dalam paper ini akan dilakukan eksperimen yaitu dengan mengintegrasikan Apache Spark yang bertindak sebagai lingkungan pemrosesan data yang besar dan konsep parallel komputing akan dikombinasikan dengan library H2O yang khusus untuk menangani pemrosesan data menggunakan algoritme machine learning. Berdasarkan hasil pengujian Apache Spark di dalam lingkungan cloud computing, Apache Spark mampu memproses data cuaca yang didapatkan dari arsip data cuaca terbesar yaitu yaitu data NCDC dengan ukuran data sampai dengan 6GB. Data tersebut diproses menggunakan salah satu model machine learning yaitu deep learning dengan membagi beberapa node yang telah terbentuk di lingkungan cloud computing dengan memanfaatkan library H2O. Keberhasilan tersebut dapat dilihat dari parameter pengujian yang telah diujikan meliputi nilai running time, throughput, Avarege Memory dan Average CPU yang didapatkan dari Benchmark Hibench. Semua nilai tersebut  dipengaruhi oleh banyaknya data dan jumlah node.</p><p class="Judul2"> </p><p class="Judul2"><em><strong>Abstract</strong></em></p><p><em>Apache Spark is a platform that can be used to process data with relatively large data sizes (big data) with the ability to divide the data into each cluster that has been determined. This concept is called parallel computing. Apache Spark has advantages compared to other similar frameworks such as Apache Hadoop, etc., where Apache Spark is able to process data in streaming, meaning that the data entered into the Apache Spark environment can be directly processed without waiting for other data to be collected. In order for Apache Spark to be able to do machine learning processes, in this paper an experiment will be conducted that integrates Apache Spark which acts as a large data processing environment and the concept of parallel computing will be combined with H2O libraries specifically for handling data processing using machine learning algorithms . Based on the results of testing Apache Spark in a cloud computing environment, Apache Spark is able to process weather data obtained from the largest weather data archive, namely NCDC data with data sizes up to 6GB. The data is processed using one of the machine learning models namely deep learning by dividing several nodes that have been formed in the cloud computing environment by utilizing the H2O library. The success can be seen from the test parameters that have been tested including the value of running time, throughput, Avarege Memory and CPU Average obtained from the Hibench Benchmark. All these values are influenced by the amount of data and number of nodes.</em><em></em></p><p class="Judul2"><em><strong><br /></strong></em></p>

Author(s):  
. Monika ◽  
Pardeep Kumar ◽  
Sanjay Tyagi

In Cloud computing environment QoS i.e. Quality-of-Service and cost is the key element that to be take care of. As, today in the era of big data, the data must be handled properly while satisfying the request. In such case, while handling request of large data or for scientific applications request, flow of information must be sustained. In this paper, a brief introduction of workflow scheduling is given and also a detailed survey of various scheduling algorithms is performed using various parameter.


Sign in / Sign up

Export Citation Format

Share Document