Data-Intensive Computing: From Clouds to GPUs Gagan Agrawal June 1, 20161

2  Growing need for analysis of large scale data  Scientific  Commercial  Data-intensive Supercomputing (DISC)  Map-Reduce has received a lot of attention  Database and Datamining communities  High performance computing community Motivation

Motivation (2)  Processor Architecture Trends  Clock speeds are not increasing  Trend towards multi-core architectures  Accelerators like GPUs are very popular  Cluster of multi-cores / GPUs are becoming common  Trend Towards the Cloud  Use storage/computing/services from a provider  How many of you prefer gmail over cse/osu account for ?  Utility model of computation  Need high-level APIs and adaptation June 1, 20163

My Research Group  Data-intensive theme at multiple levels  Parallel programming Models  Multi-cores and accelerators  Adaptive Middleware  Scientific Data Management / Workflows  Deep Web Integration and Analysis June 1, 20164

Personnel  Currently  10 PhD students  4 MS thesis students  Graduated PhDs  7 Graduated PhDs between 2005 and 2008 June 1, 20165

This Talk  Parallel Programming API for Data-Intensive Computing  An alternate API and System for Google’s Map- Reduce  Show actual comparison  Data-intensive Computing on Accelerators  Compilation for GPUs  Overview of other topics  Scientific Data Management  Adaptive and Streaming Middleware  Deep web June 1, 20166

Map-Reduce  Simple API for (data-intensive) parallel programming  Computation is:  Apply map on each input data element  Produce ( key,value ) pair(s)  Sort them using the key  Apply reduce on the set with a distinct key values June 1, 20167

8 Map-Reduce Execution

June 1,  Positives:  Simple API  Functional language based  Very easy to learn  Support for fault-tolerance  Important for very large-scale clusters  Questions  Performance?  Comparison with other approaches  Suitability for different class of applications? Map-Reduce: Positives and Questions

Class of Data-Intensive Applications  Many different types of applications  Data-center kind of applications  Data scans, sorting, indexing  More ``compute-intensive’’ data-intensive applications  Machine learning, data mining, NLP  Map-reduce / Hadoop being widely used for this class  Standard Database Operations  Sigmod 2009 paper compares Hadoop with Databases and OLAP systems  What is Map-reduce suitable for?  What are the alternatives?  MPI/OpenMP/Pthreads – too low level? June 1,

Hadoop Implementation June 1,  HDFS  Almost GFS, but no file update  Cannot be directly mounted by an existing operating system  Fault tolerance  Name node  Job Tracker  Task Tracker

June 1, FREERIDE: GOALS  Framework for Rapid Implementation of Data Mining Engines  The ability to rapidly prototype a high- performance mining implementation  Distributed memory parallelization  Shared memory parallelization  Ability to process disk-resident datasets  Only modest modifications to a sequential implementation for the above three  Developed at Ohio State June 1,

FREERIDE – Technical Basis June 1,  Popular data mining algorithms have a common canonical loop  Generalized Reduction  Can be used as the basis for supporting a common API  Demonstrated for Popular Data Mining and Scientific Data Processing Applications While( ) { forall (data instances d) { I = process(d) R(I) = R(I) op f(d) } ……. }

June 1,  Similar, but with subtle differences Comparing Processing Structure

Observations on Processing Structure  Map-Reduce is based on functional idea  Do not maintain state  This can lead to sorting overheads  FREERIDE API is based on a programmer managed reduction object  Not as ‘clean’  But, avoids sorting  Can also help shared memory parallelization  Helps better fault-recovery June 1,

June 1,  Tuning parameters in Hadoop  Input Split size  Max number of concurrent map tasks per node  Number of reduce tasks  For comparison, we used four applications  Data Mining: KMeans, KNN, Apriori  Simple data scan application: Wordcount  Experiments on a multi-core cluster  8 cores per node (8 map tasks) Experiment Design

June 1,  KMeans: varying # of nodes Avg. Time Per Iteration (sec) # of nodes Dataset: 6.4G K : 1000 Dim: 3 Results – Data Mining

June 1, Results – Data Mining (II) June 1,  Apriori: varying # of nodes Avg. Time Per Iteration (sec) # of nodes Dataset: 900M Support level: 3% Confidence level: 9%

June 1, June 1,  KNN: varying # of nodes Avg. Time Per Iteration (sec) # of nodes Dataset: 6.4G K : 1000 Dim: 3 Results – Data Mining (III)

June 1,  Wordcount: varying # of nodes Total Time (sec) # of nodes Dataset: 6.4G Results – Datacenter-like Application

June 1,  KMeans: varying dataset size Avg. Time Per Iteration (sec) Dataset Size K : 100 Dim: 3 On 8 nodes Scalability Comparison

June 1,  Wordcount: varying dataset size Total Time (sec) Dataset Size On 8 nodes Scalability – Word Count

June 1,  Four components affecting the hadoop performance  Initialization cost  I/O time  Sorting/grouping/shuffling  Computation time  What is the relative impact of each ?  An Experiment with k-means Overhead Breakdown

June 1,  Varying the number of clusters (k) Avg. Time Per Iteration (sec) # of KMeans Clusters Dataset: 6.4G Dim: 3 On 16 nodes Analysis with K-means

June 1,  Varying the number of dimensions Avg. Time Per Iteration (sec) # of Dimensions Dataset: 6.4G K : 1000 On 16 nodes Analysis with K-means (II)

Observations June 1,  Initialization costs and limited I/O bandwidth of HDFS are significant in Hadoop  Sorting is also an important limiting factor for Hadoop’s performance

This Talk  Parallel Programming API for Data-Intensive Computing  An alternate API and System for Google’s Map- Reduce  Show actual comparison  Data-intensive Computing on Accelerators  Compilation for GPUs  Overview of other topics  Scientific Data Management  Adaptive and Streaming Middleware  Deep web June 1,

Background - GPU Computing Many-core architectures/Accelerators are becoming more popular GPUs are inexpensive and fast CUDA is a high-level language for GPU programming

CUDA Programming Significant improvement over use of Graphics Libraries But.. Need detailed knowledge of the architecture of GPU and a new language Must specify the grid configuration Deal with memory allocation and movement Explicit management of memory hierarchy

Parallel Data mining Common structure of data mining applications (FREERIDE)‏ /* outer sequential loop *//* outer sequential loop */ while() { while() { /* Reduction loop */ /* Reduction loop */ Foreach (element e){ Foreach (element e){ (i, val) = process(e); (i, val) = process(e); Reduc(i) = Reduc(i) op val; Reduc(i) = Reduc(i) op val; } }

Porting on GPUs  High-level Parallelization is straight-forward  Details of Data Movement  Impact of Thread Count on Reduction time  Use of shared memory

Architecture of the System Variable information Reduction functions Optional functions Code Analyzer( In LLVM) ‏ Variable Analyzer Code Generator Variable Access Pattern and Combination Operations Host Program Grid configuration and kernel invocation Kernel functions Executable User Input

A sequential reduction function Optional functions (initialization function, combination function…) ‏ Values of each variable or size of array Variables to be used in the reduction function

Analysis of Sequential Code Get the information of access features of each variable Determine the data to be replicated Get the operator for global combination Variables for shared memory

Memory Allocation and Copy Copy the updates back to host memory after the kernel reduction function returns C.C.C.C. Need copy for each thread T0T1 T2 T3 T4 T61T62 T63T0T1 …… T0T1 T2T3T4 T61T62 T63T0T1 …… A.A.A.A. B.B.B.B.

Generating CUDA Code and C++/C code Invoking the Kernel Function Memory allocation and copy Thread grid configuration (block number and thread number) ‏ Global function Kernel reduction function Global combination

Optimizations Using shared memory Providing user-specified initialization functions and combination functions Specifying variables that are allocated once

Applications K-means clustering EM clustering PCA

Experimental Results Speedup of k-means

Speedup of k-means on GeForce 9800X2

Speedup of EM

Speedup of PCA

Deep Web Data Integration  The emerge of deep web  Deep web is huge  Different from surface web  Challenges for integration  Not accessible through search engines  Inter-dependences among deep web sources

Our Contributions  Structured Query Processing on Deep Web  Schema Mining/Matching  Analysis of Deep web data sources P44

45 Adaptive Middleware  To Enable the Time-critical Event Handling to Achieve the Maximum Benefit, While Satisfying the Time/Budget Constraint  To be Compatible with Grid and Web Services  To Enable Easy Deployment and Management with Minimum Human Intervention  To be Used in a Heterogeneous Distributed Environment ICAC 2008

46 HASTE Middleware Design ICAC 2008

47 Workflow Composition System

Summary  Growing data is creating new challenges in HPC, grid and cloud environments  Number of topics being addressed  Many opportunities for involvement  888 meets Thursdays 5:00 – 6:00, DL 280 June 1,