pact 2010 big data workloads an architect’s perspective lizy k. john university of texas at austin...

39
PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

Upload: june-bailey

Post on 17-Dec-2015

216 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

Big Data WorkloadsAn Architect’s Perspective

Lizy K. JohnUniversity of Texas at Austin

BPOE 2014

Page 2: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

The Buzz with Big Data

3/1/2014

BPOE 2014

Lizy K. John

Page 3: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

BIG DATA - Seeing things we could not see before

3/1/2014

BPOE 2014

Lizy K. John

Analyze massive amounts of data

Derive Insights

Business

Medicine

World Economy

Page 4: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

3/1/2014

BPOE 2014

An Architect would like to know

Lizy K. John

What kind of cores, memory organizations and clustering support needed to support big data

Performance metrics to guide workload partitioning strategies other than use available/affordable nodes

Partitioning considering performance, power, energy

Scaling of computation and communication depending on partitions

Becomes important to understand big data workloads

Page 5: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

Common Definition

“Data that is too large and complex to classify using traditional relational database methods”

-Wikipedia

3/1/2014

BPOE 2014

What is “Big Data”?

Lizy K. John

1 Terabyte?

– Yesterdays “Big Data”

Petabytes? Exabytes?

– Today’s “Big Data”

Zettabytes?

– Tomorrow’s “Big Data”

What does complex mean??

Need a more complete definition

Page 6: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

3/1/2014

BPOE 2014

Some examples

Lizy K. John

Combined Space of all hard drives in 2006

– 160 exabytes

All hard drives sold by Seagate in 2011

– 300 exabytes

The world wide web in 2013

– 4 zettabytes

NSA Utah Data Center in Snowden leaks

– 5 zettabytes (some claimed it to be 1 YB)

Exa = 2^60

Zetta = 2^70

Yotta = 2^80

Page 7: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

3/1/2014

BPOE 2014

Characteristics of Big Data

Lizy K. John

* Not always included in taxonomy

Page 8: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

3/1/2014

BPOE 2014

Big Data Analytics = I got this in the mail the very same week my son turned 16

Lizy K. John

Page 9: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

3/1/2014

BPOE 2014

What’s the Problem?

Lizy K. John

Deriving insights from data NOT a new problem

– Traditional relational databases that contain carefully pruned and organized data

But storage is relatively cheap these days

– Possible to store more data in unstructured form

Need intelligent ways to distill large amounts of data in different formats to actionable KNOWLEDGE

Many different levels to approach this problem…..

Page 10: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

3/1/2014

BPOE 2014

Big Data Stack

Lizy K. John

Algorithms

– PageRank, Genetic Algorithms, SVM, etc.

Frameworks and Implementations

– Map/Reduce (Hadoop), MySQL, NoSQL (Cassandra), etc

Hardware

– SMT, Accelerator Nodes (Intel Phi, GPU), etc

How does workload analysis fit in?

– EVERYONE BENEFITS FROM A DEEP UNDERSTANDING OF A WORKLOAD AND ITS CHARACTERISTICS!

Page 11: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

3/1/2014

BPOE 2014

Are New Benchmarks Needed?

Lizy K. John

Already have industry standard benchmarks!

Critical Question

– Do Big Data workloads have different characteristics than these “traditional” Benchmarks?

– Yes they do!

• TLB Behavior [Wang et al]• I-Cache Behavior [Ferdman et al, Zhen et al, Wang et al]• SMT [Ferdman et al]• Operation Intensity [Wang et al]• Data Volume [Wang et al]

Page 12: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

3/1/2014

BPOE 2014

Why New Benchmarks?

Lizy K. John

I-Cache behavior from Cloudsuite [Ferdman et al]

– Much higher miss rate than traditional benchmarks

– Significant OS contribution to cache behavior

Page 13: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

3/1/2014

BPOE 2014

Why New Benchmarks?

Lizy K. John

OS Activity [Zhen et al]

– Shows percentage of instructions

– Significant variation in kernel/application dynamic instructions

Page 14: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

3/1/2014

BPOE 2014

Why New Benchmarks?

Lizy K. John

I-TLB Behavior from BigDataBench [Wang et al]

– Once again, more misses than traditional benchmarks

Page 15: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

3/1/2014

BPOE 2014

Big Data Characterization Challenges

Lizy K. John

INPUT GENERATION

Input data is critical!

Couple of approaches

– Synthetic data generation • Questionable Veracity

– Grab data from industry• Not always possible

• CAIDA-like

How much data?

– Feasibility vs accuracy

Page 16: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

3/1/2014

BPOE 2014

Common Big Data Domains

Lizy K. John

Databases

– Structured

• Typically relational data• SQL databases

– Unstructured

• Example: document oriented• Generally no fixed table schema

– Semi-structured

Page 17: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

3/1/2014

BPOE 2014

Common Big Data Domains

Lizy K. John

Common NoSQL Databases

– Cassandra

• Industry leading, ultra scalable

– HBase

• Database built on top of Hadoop and HDFS

– MongoDB

• JSON- database with dynamic schema

Page 18: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

3/1/2014

BPOE 2014

Common Big Data Domains

Lizy K. John

Map/Reduce - Hadoop

Key/ Value computation

– Map and Reduce phase

Page 19: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

3/1/2014

BPOE 2014

Common Big Data Domains

Lizy K. John

Graph Algorithms

– Important for Data Mining and Machine Learning

– Graphlab – essentially Hadoop over large graphs

– GraphChi – web scale graph computation

– Streaming graph changes

– asynchronous changes to the graph  (i.e changes written to edges are immediately visible to subsequent computation)

– Partitioning Challenges

Page 20: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

3/1/2014

BPOE 2014

Hierarchical Decomposition of Workloads

Lizy K. John

By dividing into functional blocks - e.g. front end, back end, and database.

By subdividing into tasks, task groups, processes, threads, etc.

By dividing considering hardware modules at microarchitectural level – memory subsystem, CPU, disk, etc. eg: consider AMD APUs

Group together tasks in an application that use data from the same rack.

Page 21: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

3/1/2014

BPOE 2014

Entropy Guided Optimizations

Lizy K. John

Partitioning Graph Workloads

– How do we assign work to nodes?

Important Factors

– Data Locality

– Minimize Communication

– Maximize Resource Utilization

Bisection bandwidth

Entropy Guided Optimization

Entropy = (memory-in, memory out, #computations, …other attributes)

Page 22: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

3/1/2014

BPOE 2014

In-Memory Map/Reduce

Lizy K. John

IBM Main Memory Map Reduce (M3R)

– Eliminates intermediate disk writes for Hadoop Map/Reduce Jobs

– Pros

• Significantly speeds up some workloads– 45x on sparse matrix mult

– Cons

• Data must fit in cluster memory

• No failure resilience

Page 23: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

3/1/2014

BPOE 2014

Big Data Benchmarking Challenges

Lizy K. John

WORKLOAD VARIETY

Ton of software stacks required

– Configuration of software platform sometimes more important than workload (see next slide)

A comprehensive benchmark should feature

– Offline (Batch Style Analytics)

– Online (Real Time Analytics)

Seeing positive momentum here!

TPC-* -> Cloudsuite, BigDataBench, etc

Page 24: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

3/1/2014

BPOE 2014

Hadoop Case Study – Optimal Settings

Lizy K. John

What are the optimal framework settings?

– Workload Dependent?

– Hardware Dependent?

– Just set everything to the maximum value??

– Does it matter?

How do engineers setup clusters for new platforms?

– Some “rules of thumb” available, but imprecise

Page 25: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

3/1/2014

BPOE 2014

Hadoop Case Study

Lizy K. John

Standard Hadoop configuration algorithm ):-

hadoop_options = Google(“Best Hadoop Configuration”)

launch_cluster()

if (!cluster_boots || !clients_happy) {

hadoop_options = Permute(hadoop_options)

launch_cluster()

if(!cluster_boots || !clients_happy) {

options = Lookup_Options(Buddy_at_Other_Company)

launch_cluster()

if(!cluster_boots || !clients_happy) {

options = default_options

launch_cluster()

}

}

Page 26: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

3/1/2014

BPOE 2014

Hadoop Case Study (Mapper-Reducer Slots)

Lizy K. John

16m4r

2m2r

32m4r8m8r

CPU Occupancy of TeraSort for different mapper-reducer slots

– Simple app, but different very different execution profile depending on configuration

64m4r

Page 27: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

3/1/2014

BPOE 2014

Hadoop Case Study (Mapper-Reducer Slots)

Lizy K. John

Memory Utilization of TeraSort for different mapper-reducer slots

– Simple app, but different very different execution profile depending on configuration

16m4r2m2r 32m4r8m8r

64m4r

Page 28: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

3/1/2014

BPOE 2014

Hadoop Case Study (Block Size)

Lizy K. John

TeraSort – Higher block size reduces total number of maps

– Simple app, but different very different execution profile depending on configuration

32MB 64MB 128MB 256MB 512MB

Page 29: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

3/1/2014

BPOE 2014

Big Data Benchmarking Frameworks

Lizy K. John

Management frameworks and harnesses essential

Example: AMD SWAT

– Software platform for automating the…..

• creation, deployment, provisioning, execution, and data gathering of synthetic workloads on scalable clusters

Several benchmarks available

– Cloudsuite

– Hadoop

– Graphlab

– Anything you want to plugin!

Page 30: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

Lizy K. John 3/1/2014

Page 31: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

3/1/2014

BPOE 2014

Big Data Benchmarking Challenges

Lizy K. John

Big Cluster

Lots of cores, lots of memory and disk space

– Hard for non-industry researchers

Prohibitively long runtimes

Can we simulate Big Data?• Requires full system simulation• Cloudsuite on Flexus (EPFL)

Page 32: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

3/1/2014

BPOE 2014

Adaptable Scalable Futuristic Benchmark Proxies

Lizy K. John

Generate Clones by setting knobs to appropriate values

Adaptable

Scalable

Futuristic

Benchmark Synthesizer

Application Behavior Space

‘Knobs’ for Changing Program

Characteristcs

Workload Synthesis Algorithm

Synthetic Benchmark

Pre-silicon ModelHardwareCompile and Execute

Workload Characteristics

Page 33: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

3/1/2014

BPOE 2014

Lizy K. John

Abstract

Workload

Model

No. Metric Category

1 Basic block size Control flow predictability 2 Branch taken rate for each branch

3 Branch transition rate

4 Proportion of INT ALU, INT MUL, INT DIV, FP ADD, FP MUL, FP DIV, FP MOV, FP SQRT, LOAD & STORE

Instruction mix

5 Dependency distance distribution Instruction level parallelism

6 Private stride value per static load/storeData locality

7 Data Footprint of the workload

8 Mean and standard deviation of the MLP Memory Level Parallelism (MLP)9 MLP frequency

10 Number of threads Thread level parallelism

11 Thread class and processor assignment

Shared data access pattern and communication characteristics

12 Percentage loads to private data

13 Percentage loads to read-only data

14 Percentage migratory loads

15 Percentage consumer loads

16 Percentage irregular loads

17 Percentage stores to private data

18 Percentage producer stores

19 Percentage irregular stores

20 Shared stride value per static load/store

21 Data pool distribution based on sharing patterns

22 Number of lock/unlock pairs andSynchronization Characteristics

23 Number of mutex objects

24 Number of Instructions between lock and unlock

Page 34: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

3/1/2014

BPOE 2014

Big Data Synthetics? A Possibility?

Lizy K. John

Given challenges in Big Data workloads, this would be useful

But what are the knob settings for “Big Data”

– Need detailed characterization

Benchmark Synthesizer

Program

LocalityInstru

ction

Mix Control F

low

BehaviorApplication

Behavior Space

‘Knobs’ for Changing Program

Characteristcs

Workload Synthesis Algorithm

Synthetic Benchmark

Pre-silicon ModelHardwareCompile and Execute

Workload Characteristics

Thread Level

ParallelismCommunicatio

n

characteristic

s

Data Sharing

Patterns

Page 35: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

3/1/2014

BPOE 2014

Big Data Workload Clones

Lizy K. John

CLONES WILL AVOID COMPLEX SOFTWARE STACKS:

Clones for Hadoop

Clones for Graph Processing

Clones for DSS

Clones for OLAP

Clones for DSS with materialized views

Need detailed characterization

Benchmark Synthesizer

Program

LocalityInstru

ction

Mix Control F

low

BehaviorApplication

Behavior Space

‘Knobs’ for Changing Program

Characteristcs

Workload Synthesis Algorithm

Synthetic Benchmark

Pre-silicon ModelHardwareCompile and Execute

Workload Characteristics

Thread Level

ParallelismCommunicatio

n

characteristic

s

Data Sharing

Patterns

Page 36: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

3/1/2014

BPOE 2014

Tricks from the Old Treasure Chest

Lizy K. John

Search and Sort –

– age old computer science problems

– new issues raised by scale but

Old OLTP, OLAP and DSS

Combination of HPC and Database Ideas

Old Scatter-Gather

Piece-wise modeling

Page 37: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

3/1/2014

BPOE 2014

Conclusion

Lizy K. John

Big Data is here to stay

Increasingly important

Cloud and Big Data will take

the world in unprecedented ways

Appropriate hardware and software need to be developed

Workload metrics to guide partitioning

Need to act now to develop intelligent benchmarks and workload analysis methodology

Page 38: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

3/1/2014

BPOE 2014

Thank You! Questions?

Laboratory for Computer Architecture (LCA)The University of Texas at Austin

lca.ece.utexas.eduLizy K. John

Page 39: PACT 2010 Big Data Workloads An Architect’s Perspective Lizy K. John University of Texas at Austin BPOE 2014

PACT 2010

[1] M. Ferdman, et. al.. 2012. Clearing the clouds: a study of emerging scale-out workloads on modern hardware.SIGARCH Comput. Archit. News 40, 1 (March 2012), 37-48.

[2] Zhen Jia, Lei Wang, Jianfeng Zhan Lixin Zhang, Chunjie Luo. Characterizing Data Analysis Workloads in Data Centers. In Workload Characterization (IISWC), 2013 IEEE International Symposium on. IEEE.

[3] Lei Wang, Jianfeng Zhan, Chunjie Luo, Yuqing Zhu, Qiang Yang, Yongqiang He, Wanling Gao, Zhen Jia, Yingjie Shi, Shujie Zhang, Cheng Zhen, Gang Lu, Kent Zhan, Xiaona Li, and Bizhu Qiu. The 20th IEEE International Symposium On High Performance Computer Architecture (HPCA-2014), February 15-19, 2014, Orlando, Florida, USA.

[4] Huang, Shengsheng, et al. "The HiBench benchmark suite: Characterization of the MapReduce-based data analysis." Data Engineering Workshops (ICDEW), 2010 IEEE 26th International Conference on. IEEE, 2010.

[5] Cooper, Brian F., et al. "Benchmarking cloud serving systems with YCSB."Proceedings of the 1st ACM symposium on Cloud computing. ACM, 2010.

[6] GridMix [Online]. Available: https://hadoop.apache.org/docs/r1.2.1/gridmix.html. (21.10.2013).

[7] PigMix [Online]. Available: https://cwiki.apache.org/confluence/display/PIG/PigMix.(21.10.2013).

[8] PAVLO, A., PAULSON, E., RASIN, A., ABADI, D.J., DEWITT, D.J., MADDEN, S., and STONEBRAKER, M., 2009. A comparison of approaches to large-scale data analysis. In Proceedings of the 2009 ACM SIGMOD International Conference on Management of data ACM, 165-178.

[9] Transaction Processing Performance Council (Online) http://www.tpc.org/default.asp (02-13-2013)

[10] GHAZAL, A., RABL, T., HU, M., RAAB, F., POESS, M., CROLOTTE, A., and JACOBSEN, H.-A., 2013. BigBench: Towards an Industry Standard Benchmark for Big Data Analytics. In SIGMOD ACM, New York, New York, 2013, 197-1208.

[11] SUMBALY, R., KREPS, J., and SHAH, S., 2013. Linkbench: a database benchmark based on the Facebook social graph In Proceedings of the SIGMOD (New York, New Youk, USA2013), ACM, 1185-1196.

[12] Cloudsuite on Flexus[Online]. http://parsa.epfl.ch/cloudsuite/isca12-tutorial.html (02-13-2013). ISCA 2012 Tutorial

[13] Graphlab [Online]. Available: http://graphlab.com/).

[14] Shinnar, A., Cunningham, D., Saraswat, V., & Herta, B. (2012). M3R: increased performance for in-memory Hadoop jobs. Proceedings of the VLDB Endowment,5(12), 1736-1747.

[15] Nambiar, Raghunath Othayoth, and Meikel Poess. "The making of TPC-DS."Proceedings of the 32nd international conference on Very large data bases. VLDB Endowment, 2006.

[16] Breternitz, Mauricio, et al. "Cloud Workload Analysis with SWAT." Computer Architecture and High Performance Computing (SBAC-PAD), 2012 IEEE 24th International Symposium on. IEEE, 2012.

Lizy K. John 3/1/2014

References