Slide 1
Application-driven Energy-efficient Architecture Explorations
for Big DataAuthors:Xiaoyan GuRui HouKe ZhangLixin ZhangWeiping
Wang(Institute of Computing Technology,Chinese Academy of
Sciences)
Reviewed by-Siddharth Bhave(University of Washington,
Tacoma)
Big DataWhat is Big Data?
Problems with Big dataEnergy ConsumptionVelocity (Operation
latency and throughput)Volume (storing capacity)Variety
Managing Big Data ProblemsStorage
TechnologiesPartitioningMultithreadingParallel ProcessingEfficient
ArchitectureHadoop, Map Reduce, MAHOUTFind bottle
neckIntroductionBig data management at architecture level
Two architecture systemsXeon-based clusterAtom Based
(micro-server) Cluster
Comparison Based on: -Energy consumptionExecution time
MotivationEver increasing data.
Energy and Time tradeoff in Xeon and Atom based clusters.
Bottleneck by the processes of compression/decompression
Stateless data processing
MastiffMastiff - Targeted application for performance
analysis
Big data processing engine
Columnar store policy
Compression Ratio on 3 GB dataCompression Ratio on 100 GB
dataCompression Ratio on 500 GB dataMastiff0.540.530.518Hadoop
HDFS0.720.710.7Working flow of the Mastiff
MethodologyTPC-H test benchmark of queries and concurrent
data
1 TB of verification data
2 cases - data load and data query
Fluke NORMA 4000
Average cases and median results are reportedPower and
Performance EvaluationTime on Atom Cluster (30 nodes)Time on Xeon
Cluster (30 nodes)Time on Xeon Cluster (15 nodes)Data Load3.435
hours1.543 hours3.242 hoursData Query5.877 hours2.724 hours5.564
hoursTake 3 cases for time and energy consumption
31 nodes Atom Cluster (1 master node)
31 nodes Xeon Cluster (1 master node)
16 nodes Xeon Cluster (1 master node)
Energy consumption between 30-node Atom Cluster and 30-node Xeon
ClusterPower and Performance Evaluation (contd)Energy consumption
between 30-node Atom Cluster and 15-node Xeon Cluster
Power and Performance Evaluation (contd)Time Breakdown in Map
Phase
Power and Performance Evaluation (contd)Time Breakdown in Reduce
phase
Power and Performance Evaluation (contd)FindingsAtom platform
more power efficient
Data compression and decompression occupies significant
percentage.
Compression and decompression can be done in software pipeline
fashion i.e. with multiple interleavePropositionsHeterogeneous
architecture
Accelerators to perform data compression/decompression
Multiple interleaved compression/decompression
Off-chip and On-chip Accelerators
Multiple Interleaved TasksStrengthsA much needed innovative
concept
Organized well
Detailed description of energy and time investigation
Already implemented propositionsWeaknessesNot enough power
meters to monitor all nodes
2 assumptionsPower of every network router is evenly counted
towards nodesEnergy consumption of each node is similar
Results are generalized by Hadoop even if they might not be true
for every application.
Vague propsitions implementationFAWN: A Fast Array of Wimpy
NodesAuthors:
David G. AndersenJason FranklinMichael KaminskyAmar
PhanishayeeLawrence TanVijay Vasudevan(Carnegie Mellon
University)High performance, energy efficient system for
storage
Large number of small low-performance (hence wimpy) nodes with
moderate amounts of local storage
2 parts: FAWN-DS (data store) and FAWN-KV (key value)
MotivationTraditional architecture consumes too much powerI/O
bottleneck due to current storage
inabilitiesIntroductionFeaturesPairs of low powered embedded nodes
with flash storage
FAWN-DS is the backend that consists of the large number of
nodes
Each node has some RAM and flash
FAWN-KV is a consistent, replicated, highly available and high
performance key value storage systemFAWN Architecture
Efficient Data Streaming with On-chip Accelerators:
Opportunities and ChanllengesAuthors:
Rui HouLixin ZhangMichael C. HuangKun WangHubertus FrankeYi
GeXiaotao Chang(University of Rochester)MotivationTransistor
density increasing day by day
Many cores are integrated in a single die
Advantage of on-chip accelerator instead of using it as PCI
On-Chip Accelerator Architecture3 types of acceleratorsCrypto
acceleratorsDecompression acceleratorsNetwork offload
accelerator
Some common characteristics of data stream in the 3
accelerators
Optimize the power and performance of the
accelerators.FeaturesThank You