sudoers: Benchmarking Hadoop with ALOJA

Benchmarking Hadoop
with ALOJA
Oct 6, 2015
by Nicolas Poggi @ni_po
sudoers Barcelona:

About Nicolas Poggi @ni_po
Work:
Education:
Community:

Agenda
 Intro on Hadoop
 Current scenario and problematic
 ALOJA project
 Open source tools
 Benchmarking DEMO
 Results
 DEMO results online
 Open questions and comments

Intro: Hadoop design and ecosystem

Hadoop design
 Hadoop designed to solve complex data
 Structured and non structured
 With [close to] linear scalability
 Simplifying the programming model
 From MPI, OpenMP, CUDA, …
 Operates as a blackbox for data analysts
Image source: Hadoop, the definitive guide

Hadoop parameters
 > 100+ tunable parameters
 mapred.map/reduce.tasks.speculative.execution
 obscure and interrelated
 io.sort.mb 100 (300)
 io.sort.record.percent 5% (15%)
 io.sort.spill.percent 80% (95 – 100%)
 Number of Mappers and Reducers
 Rule of thumb 0.5 - 2 per CPU core

Hadoop stack for tuning
Image source: Intel® Distribution for Apache Hadoop

Hadoop highly-scalable but…
 Not a high-performance solution!
 Requires
 Design,
 Clusters, topology clusters
 Setup,
 OS, Hadoop config
 and tuning required
 Iterative approach
 Time consuming
 And extensive benchmarking!

Hadoop ecosystem
 Large and spread
 Dominated by big players
 Custom patches
 Default values not ideal
 Product claims
 Cloud vs. On-premise
 IaaS
 PaaS
 EMR, HDInsight
 Needs standardization
and auditing!
DATA

Product claims
 Needs auditing!

Too many choices?
Remote volumes
-
-
Rotational HDDs
JBODs
Large VMs
Small VMs
GbEthernet
InfiniBand
RAID
Cost
Performance
On-Premise
Cloud
And where is my system
configurationpositionedon
each of these axes?
Highavailability
Replication
+
+

Project ALOJA
 Open initiative to produce mechanisms for an
 automated characterization of cost-effectiveness
 of Big Data deployments
 Results from of a growing need of the community to
understand job execution details and create transparency
 Explore different configuration deployment options and
their tradeoffs
 Both software and hardware
 Cloud services and on-premise
 Seeks to provide knowledge, tools, and an online service
 to with which users make better informed decisions
 reduce the TCO for their Big Data infrastructures
 Guide the future development and deployment of Big Data clusters
and applications

Challenges, options, and implementation

Challenges (circa end 2013)
 Test different clusters architectures
 On-premise
 Commodity, high-end, appliance, low-power
 Cloud IaaS
 32 different VMs in Azure, similar in other
providers
 Cloud PaaS
 HDInsight, EMR, CloudBigData
 Different access level
 Full admin, user-only, request-to-install,
everything ready, queuing systems (SGE)
 Different versions
 Hadoop, JVM, Spark, Hive, etc…
 Dev environments and testing
 Big Data usually requires a cluster to
develop and test

Benchmarking vs. Production envs
 Need to compare different executions
 Not how the systems are doing now
 This is the main diff with prod products
 Dada does not change (non-OLTP)
 Temporary data for benchmarks vs. Important data
 Fast iteration vs. Reliability
 Iterates configurations vs. fixed config
 Many fast, experimental changes
 Security can be relaxed
 Management for Hadoop
 Vendor lock-in
 Lack of systems support(azure, on-prem, low-power)
 Hadoop is our use case, not the only one
 Leave no traces on the benchmarked system

Available options: (circa end 2013)
 Deployment
 jclouds
 foreman
 Puppet
 Ambari
 Config and deploy
 Ambari (hadoop only)
 Use Configuration
Management (CM)
 Puppet, chef, ansible…
 Monitoring
 Ganglia, Zabbix
 Amabari
 Cloudera Manager
 Kibana, GraphD…
 Problems
 All systems thoughfor PROD
 Not for comparison
 No Azure support
 Many different packages
 No one-fits-all solution
 Solution
 Custom implementation
 Based in simple components
 Wrapping commands

ALOJA Platform main components
2 Online Repository
•Explore results
•Execution details
•Cluster details
•Costs
•Data sharing
3 Web Analytics
•Data views and evaluations
•Aggregates
•Abstracted Metrics
•Job characterization
•Machine Learning
•Predictions and clustering
1 Big Data Benchmarking
•Deploy & Provision
•Conf Management
•Parameter selection & Queuing
•Perf counters
•Low-level instrumentation
•App logs
17
NGINX, PHP, MySQL
BASH, Unix tools, CLIs R, SQL, JS

Workflow in ALOJA
Cluster(s)
definition
• VM sizes
• # nodes
• OS, disks
• Capabilities
Execution
plan
• Start cluster
• Exec Benchmarks
• Gather results
• Cleanup
Import
data
• Convert perf metric
• Parse logs
• Import into DB
Evaluate
data
• Data views in Vagrant VM
• Or http://hadoop.bsc.es
PA and KD
•Predictive
Analytics
•Knowledge
Discovery
Historic
Repo
(in progress)

Cluster and node definitions
Clusters (Azure example) Node (Web in Rackspace)
#load AZURE defaults
source "$CONF_DIR/azure_defaults.conf"
clusterName="al-08"
numberOfNodes="8"
vmSize=“Large”
#details
vmCores="4"
vmRAM="7" #in GB
#costs
clusterCostHour="1.584"#0.176 * 9
clusterType="IaaS"
clusterDescription="A3 type VMs"
#load node defaults
source “$CONF_DIR/node_defaults.conf"
defaultProvider="rackspace"
vm_name="aloja-web"
vmSize='io1-30'
attachedVolumes="2"
diskSize="1023"
# Node roles (install functions)
extraLocalCommands="
vm_install_webserver;
vm_install_repo 'provider/rackspace';
install_ganglia_gmond;
config_ganglia_gmond 'aloja-web-rackspace' 'aloja-
web';
install_percona /scratch/attached/2/mysql;"

Commands and providers
Provisioning commands Providers
 Connect
 Node and Cluster
 Uses SSH proxies
automatically
 Deploy
 Start, Stop
 Delete
 Nodes and clusters
 On-premise
 Custom settings for
clusters
 Multiple disk types
 Different architectures
 Cloud IaaS
 Azure, OpenStack,
Rackspace, AWS (testing)
 Cloud PaaS
 HDInsight, CloudBigData,
EMR soon
Code at: https://github.com/Aloja/aloja/tree/master/aloja-deploy

Running benchmarks in ALOJA
 Example of submitting a job to run:
 https://github.com/Aloja/aloja/blob/master/aloja-bench/run_benchs.sh
 To queue jobs and control results:
 https://github.com/Aloja/aloja/blob/master/shell/exeq.sh

ALOJA Online Benchmark Repository
 Entry point for explorethe results collected from the executions
 Index of executions
 Quick glance of executions
 Searchable,Sortable
 Execution details
 Performance chartsandhistograms
 Hadoopcounters
 Jobsand taskdetails
 Data management of benchmark executions
 Data importing from different clusters
 Execution validation
 Data management and backup
 Cluster definitions
 Cluster capabilities (resources)
 Cluster costs
 Sharing results
 Download executions
 Add external executions
 Documentation and References
 Papers, links, and feature documentation
Availableat: http://aloja.bsc.es

Impact of SW configurations in Speedup
(4 node clusters)
Number of mappers Compression algorithm
No comp.
ZLIB
BZIP2
snappy
4m
6m
8m
10m
Speedup (higher is better)
Results using: http://hadoop.bsc.es/configimprovement
Details: https://raw.githubusercontent.com/Aloja/aloja/master/publications/BSC-MSR_ALOJA.pdf

Impact of HW configurationsin Speedup
Disks and Network Cloud remote volumes
Local only
1 Remote
2 Remotes
3 Remotes
3 Remotes
/tmp local
2 Remotes
/tmp local
1 Remotes
/tmp local
HDD-ETH
HDD-IB
SSD-ETH
SDD-IB
Speedup (higher is better)
Results using: http://hadoop.bsc.es/configimprovement
Details: https://raw.githubusercontent.com/Aloja/aloja/master/publications/BSC-MSR_ALOJA.pdf

Speedup: all disk configurationsSSD vs JBOD
 For DFSIOEread, DFSIOEwrite, and Terasort
URL:
http://hadoop.bsc.es/configimprovement?datefrom=&dateto=&benchs%5B%5D=dfsioe_read&benchs%5B%5D=dfsioe_write&benchs%5B%5D=terasort&id_clusters%5B%5D=21&nets%5B%5D=None&disks%5B%5D=HD2&disks%5B%5D=H
D3&disks%5B%5D=HD4&disks%5B%5D=HD5&disks%5B%5D=HDD&disks%5B%5D=HS5&disks%5B%5D=RL1&disks%5B%5D=RL2&disks%5B%5D=RL3&disks%5B%5D=RL4&disks%5B%5D=RL5&disks%5B%5D=RL6&disks%5B%5D=RR1&disks%
5B%5D=SS2&disks%5B%5D=SSD&mapss%5B%5D=None&comps%5B%5D=None&replications%5B%5D=None&blk_sizes%5B%5D=None&iosfs%5B%5D=None&iofilebufs%5B%5D=None&datanodess%5B%5D=None&bench_types%5B%5D=H
DI&bench_types%5B%5D=HiBench&vm_sizes%5B%5D=None&vm_coress%5B%5D=None&vm_RAMs%5B%5D=None&hadoop_versions%5B%5D=None&types%5B%5D=None&filters%5B%5D=valid&filters%5B%5D=filters&allunchecked=
2 SSDs
5 SATA
1 SSD /tmp
1 SSD
1 SATA
2 SATA
3 SATA
4 SATA
5 SATA
Higherisbetter
Fastest config
Highcapacity and fast
Highcapacity but slow

Speedup by disk configuration in the Cloud
(higher is better)
URL
http://104.130.159.92/configimprovement?benchs%5B%5D=terasort&disks%5B%5D=HDD&disks%5B%5D=RL1&disks%5B%5D=RL2&disks%5B%5D=RL3&disks%5B%5D=RR1&disks%5B%5D=RR2&disk
s%5B%5D=RR3&disks%5B%5D=RR4&disks%5B%5D=RR5&disks%5B%5D=RR6&disks%5B%5D=RS1&disks%5B%5D=RS6&disks%5B%5D=SSD&bench_types%5B%5D=HiBench&filters%5B%5D=valid&filt
ers%5B%5D=filters&allunchecked=&selected-groups=disk&datefrom=&dateto=&minexetime=150&maxexetime=1500
1-6 remotes
1 and 6
remotes with
/tmp on SSD
SSD only
Higherisbetter

VM Size comparison(Azure) Lower is better

Preview: Cost/Performance Scalability
 This shows a sample of a new screen (with sample data) to find the most cost-
effective cluster size
 X axis number of datanodes (cluster size
 Left Y Execution time (lower is better)
 Right Y Execution cost
Execution time Execution cost
Recommendedsize

InfiniBand + SDD (LOCAL)
GbE SDD + (LOCAL) CLOUD (local disk/tmpand HDFS)
CLOUD (/tmpinLocal Disk, HDFSin Blob storage 1-3
devices)
CLOUD (/tmpandHDFSin Blob storage
1-3 devices)
InfiniBand + SATA disks (LOCAL)
GbE+ SATA disks (LOCAL)
Price
Performance
Cost-effectiveness On-premise vs. Cloud)
Details at: https://raw.githubusercontent.com/Aloja/aloja/master/publications/BSC-MSR_ALOJA.pdf

Open questions:
is BASH good enough?
PROs CONs and Alternatives
 Simple and Fast
 Well known
 (basics at least)
 Easy to hack
 Most of the work
requires running sys
commands
 Custom implementation
problems
 Missing some systems
 Too simple, missing:
 objects, inheritance,
types, data structures,
testing
 Python? Perl?
 Puppet? Ansible?
 We’ll stick to bash for
now..
 What’s missing for
incubating in Apache?

More info:
 ALOJA Benchmarking platform and online repository
 http://aloja.bsc.es
 Benchmarking Big Data by Nicolas Poggi
 http://www.slideshare.net/ni_po/benchmarking-hadoop
 Big Data Benchmarking Community (BDBC) mailing list
 (~200 members from ~80organizations)
 http://clds.sdsc.edu/bdbc/community
 Workshop Big Data Benchmarking (WBDB)
 Next: http://clds.sdsc.edu/wbdb2015.ca
 SPEC Research Big Data working group
 http://research.spec.org/working-groups/big-data-working-group.html
 Slides and video:
 Michael Frank on Big Data benchmarking
 http://www.tele-task.de/archive/podcast/20430/
 Tilmann Rabl BigData BenchmarkingTutorial
 http://www.slideshare.net/tilmann_rabl/ieee2014-tutorialbarurabl

@BDOOP_BCN
More info: http://aloja.bsc.es
or join BDOOP group
http://www.meetup.com/Barcelona-BigData-Perfomance-and-
Operations
Oct 06, 2015

sudoers: Benchmarking Hadoop with ALOJA

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to sudoers: Benchmarking Hadoop with ALOJA (20)

Recently uploaded (20)

sudoers: Benchmarking Hadoop with ALOJA