Platforms to handle Big data

Dr. Jigna Ashish Patel

Assistant Professor, CSE Dept,
Institute of Technology,
Nirma University
Objective of the lecture
• Right platform
• Need of the application/algorithm
• Right decision
Application/algorithm level requirements
• How quickly do we need to get the results?
• How big is the data to be processed?
• Does the model building require several iterations or single iteration?
System/platform level requirements
• Will there be a need for more data processing capability in the
future?
• Is the rate of data transfer critical for this application?
• Is there a need for handling hardware failures within the application?
Scaling
• Scaling is the ability of the system to adapt to increased demands in
terms of data processing
Horizontal Scaling
• It involves distributing the workload across many servers which may
be even commodity machines.
• It is also known as “scale out”, where multiple independent machines
are added together in order to improve the processing capability.
• Typically, multiple instances of the operating system are running on
separate machines.
Vertical Scaling
• Vertical Scaling involves installing more processors, more memory and
faster hardware, typically, within a single server.
• It is also known as “scale up” and it usually involves a single instance
of an operating system.
Comparison of Pros and cons
• Horizontal Scaling
Comparison of Pros and cons
• Vertical Scaling
Horizontal Scaling Platforms
• Peer-to-Peer Network
• Apache Hadoop
• Apache Spark

Vertical Scaling Platforms

• High performance computing clusters
• Multicore CPU
• Graphics Processing Unit(GPU)
• Field Programmable gate arrays(FPGA)
Peer-to-Peer networks
• involve millions of machines connected in a network
• decentralized and distributed network architecture where the
nodes in the networks (known as peers) serve as well as consume
resources.
• oldest distributed computing platforms
• Message Passing Interface (MPI) for communication scheme used
in such a setup to communicate and exchange the data between
peers.
• Each node can store the data instances and the scale out is
practically unlimited (can be millions of nodes).
Apache Hadoop
• open source framework for storing and processing large
datasets using clusters of commodity hardware.
• Hadoop is designed to scale up to hundreds
• highly fault tolerant
• The Hadoop platform contains the following two important
components: (1) HDFS (2) YARN
Apache Spark
• developed by researchers at the University of California at
Berkeley. designed to overcome the disk I/O limitations
• ability to perform in-memory computations.
• allows the data to be cached in memory, thus eliminating the
Hadoop’s disk overhead limitation for iterative tasks.
• supports Java, Scala and Python and for certain tasks
• it is tested to be up to 100× faster than Hadoop MapReduce
HPC clusters
• Known as blades or supercomputers, are machines with thousands
of cores.
• They can have a different variety of disk organization, cache,
communication mechanism etc.
• powerful hardware which is optimized for speed and throughput.
• They are not as scalable as Hadoop or Spark clusters but they are
still capable of processing terabytes of data.
Multicore CPU
• Multicore refers to one machine having dozens of processing
cores They usually have shared memory but only one disk.
• the number of cores per chip and the number of operations that
a core can perform has increased significantly. Newer breeds of
motherboards allow multiple CPUs within a single machine
thereby increasing the parallelism.
• Until the last few years, CPUs were mainly responsible for
accelerating the algorithms for big data analytics.
GPU
• It is designed to accelerate the creation of images in a frame
buffer intended for display output
• GPUs were primarily used for graphical operations such as video
and image editing, accelerating graphics-related processing etc.
due to their massively parallel architecture, recent
developments in GPU hardware and related programming
frameworks have given rise to GPGPU
• In addition to the processing cores, GPU has its own high
throughput DDR5 memory which is many times faster than a
typical DDR3 memory.
FPGA
• highly specialized hardware units for specific applications
• FPGAs can be highly optimized for speed and can be orders of
magnitude faster compared to other platforms for certain
applications.
• Due to customized hardware, the development cost is typically
much higher compared to other platforms.
• On the software side, coding has to be done in HDL with a low-
level knowledge of the hardware which increases the algorithm
development cost.
Comparison of platforms
System/Platform Level characteristics

• Scalability
• Data I/O performance
• Fault Tolerance
Scalability

Platform Scalability
Peer-to-Peer *****
Virtual Clusters(MapReduce/MPI) *****
Virtual Clusters(Spark) *****
HPC clusters (MPI/MapReduce) ***
Multicore(Multithreading) **
GPU(CUDA) **
FPGA(HDL) *
Data I/O Performance

Platform Data I/O performance

Peer-to-Peer *
Virtual Clusters(MapReduce/MPI) **
Virtual Clusters(Spark) ***
HPC clusters (MPI/MapReduce) ****
Multicore(Multithreading) ****
GPU(CUDA) *****
FPGA(HDL) *****
Fault Tolerance

Platform Fault Tolerance

Peer-to-Peer *
Virtual Clusters(MapReduce/MPI) *****
Virtual Clusters(Spark) *****
HPC clusters (MPI/MapReduce) ****
Multicore(Multithreading) ****
GPU(CUDA) ****
FPGA(HDL) ****
Comparison of platforms
Application/Algorithm Level characteristics

• Real time processing

• Data size supported
• Iterative task support
Real Time Processing

Platform Real Time Processing

Peer-to-Peer *
Virtual Clusters(MapReduce/MPI) **
Virtual Clusters(Spark) **
HPC clusters (MPI/MapReduce) ***
Multicore(Multithreading) ***
GPU(CUDA) *****
FPGA(HDL) *****
Data Size supported

Platform Data Size supported

Peer-to-Peer *****
Virtual Clusters(MapReduce/MPI) ****
Virtual Clusters(Spark) ****
HPC clusters (MPI/MapReduce) ****
Multicore(Multithreading) **
GPU(CUDA) **
FPGA(HDL) **
Iterative Task Support

Platform Iterative task support

Peer-to-Peer **
Virtual Clusters(MapReduce/MPI) **
Virtual Clusters(Spark) ***
HPC clusters (MPI/MapReduce) ****
Multicore(Multithreading) ****
GPU(CUDA) ****
FPGA(HDL) ****
How will you choose one of platform for a
particular criteria ?
Amount of Time
Number of Iterations
Fault Tolerance
Scalability
Choice of platform
• Data size
• Speed/Throughput
• Training /Applying a model
K means clustering
K-means on MapReduce
K-means on MPI
K-means on GPU