This document discusses performance optimization of Apache Accumulo, a distributed key-value store. It describes modeling Accumulo's bulk ingest process to identify bottlenecks, such as disk utilization during the reduce phase. Optimization efforts included improving data serialization to speed sorting, avoiding premature data expansion, and leveraging compression. These techniques achieved a 6x speedup. Current Accumulo performance projects include optimizing metadata operations and write-ahead log performance.

1. Securely explore your data
PERFORMANCE MODELS
FOR APACHE ACCUMULO:
THE HEAVY TAIL OF A SHARED-NOTHING
ARCHITECTURE
Chris McCubbin
Director of Data Science
Sqrrl Data, Inc.

2. I’M NOT ADAM FUCHS
• But perhaps I’m still an interesting guy
• MS in CS from UMBC in Network Security and
Quantum Computing
• 8 years at JHU/APL working on UxV Swarms
• 4 years at JHU/APL and TexelTek creating Big
Data Applications for the NSA
• Co-founder and Director of Data Science at Sqrrl
©2014 Sqrrl Data, Inc 2

3. SO, YOUR DISTRIBUTED
APPLICATION IS SLOW
• Today’s distributed applications run on tens or
hundreds of library components
• Many versions so internet advice could be ineffective, or
worse, flat out wrong
• Hundreds of settings
• Some, shall we say, could be better documented
• Shared-nothing architectures are usually “shared-little”
architectures with tricky interactions
• Profiling is hard and time-consuming
• What do we do?
©2014 Sqrrl Data, Inc 3

4. TODAY’S TALK
1. Quick intro to performance optimization
2. Tricks and techniques for targeted distributed
application modeling performance improvement
3. A deep dive into improving bulk load application
performance
©2014 Sqrrl Data, Inc 4

5. The Apache Accumulo™ sorted, distributed key/value store is a secure, robust,
scalable, high performance data storage and retrieval system.
• Many applications in real-time storage and analysis of “big data”:
• Spatio-temporal indexing in non-relational distributed databases - Fox et al
2013 IEEE International Congress on Big Data
• Big Data Dimensional Analysis - Gadepally et al IEEE HPEC 2014
• Leading its peers in performance and scalability:
• Achieving 100,000,000 database inserts per second using Accumulo and
D4M - Kepner et al IEEE HPEC 2014
• An NSA Big Graph experiment (Technical Report NSA-RD-2013-056002v1)
• Benchmarking Apache Accumulo BigData Distributed Table Store Using Its
Continuous Test Suite - Sen et al 2013 IEEE International Congress on Big
Data
For more papers and presentations, see http://accumulo.apache.org/papers.html
©2014 Sqrrl Data, Inc 5

6. SCALING UP: DIVIDE & CONQUER
• Collections of KV pairs form Tables
• Tables are partitioned into Tablets
• Metadata tablets hold info about
other tablets, forming a 3-level
hierarchy
• A Tablet is a unit of work for a
Tablet Server
Table:
Adam’s
Table
Table:
Encyclopedia
Table:
Foo
Data
Tablet
-­‐∞
:
thing
Data
Tablet
thing
:
∞
Data
Tablet
-­‐∞
:
Ocelot
Data
Tablet
Ocelot
:
Yak
Data
Tablet
Yak
:
∞
Data
Tablet
-­‐∞
to
∞
Well-­‐Known
Loca9on
(zookeeper)
Root
Tablet
-­‐∞
to
∞
Metadata
Tablet
2
“Encyclopedia:Ocelot”
to
∞
Metadata
Tablet
1
-­‐∞
to
“Encyclopedia:Ocelot”
©2014 Sqrrl Data, Inc 6

7. PERFORMANCE ANALYSIS CYCLE
Simulate &
Experiment
Modify
Code
Analyze
Start:
Create
Model
Refine
Model
Outputs:
Better Code
+ Models
©2014 Sqrrl Data, Inc 7

8. MAKING A MODEL
• Determine points of low-impact metrics
• Add some if needed
• Create parallel state machine models with
components driven by these metrics
• Estimate running times and bottlenecks from
a-priori information and/or apply measured
statistics
• Focus testing on validation of the initial
model and the (estimated) pain points
• Apply Amdahl’s Law
• Rinse, repeat
©2014 Sqrrl Data, Inc 8

9. BULK INGEST OVERVIEW
• Accumulo supports two mechanisms to bring
data in: streaming ingest and bulk ingest.
• Bulk Ingest
• Goal: maximize throughput without constraining
latency.
• create a set of Accumulo Rfiles, then register those
files with Accumulo.
• RFiles are groups of sorted key-value pairs with
some indexing information
• MapReduce has a built-in key sorting phase: a good
fit to produce RFiles
©2014 Sqrrl Data, Inc 9

10. BULK INGEST MODEL
10
Map Reduce Register
Time
©2014 Sqrrl Data, Inc

11. BULK INGEST MODEL
11
Hypothetical Resource Usage
Time
• 100% CPU
• 20% Disk
• 0% Network
• 46 seconds
• 40% CPU
• 100% Disk
• 20% Network
• 168 seconds
• 10% CPU
• 20% Disk
• 40% Network
• 17 seconds
©2014 Sqrrl Data, Inc
Map Reduce Register

12. INSIGHT
• Spare disk here, spare CPU there – can we even out resource consumption?
• Why did reduce take 168 seconds? It should be more like 40 seconds.
• No clear bottleneck during registration – is there a synchronization or
serialization problem?
12
Time
• 100% CPU
• 20% Disk
• 0% Network
• 46 seconds
• 40% CPU
• 100% Disk
• 20% Network
• 168 seconds
• 10% CPU
• 20% Disk
• 40% Network
• 17 seconds
©2014 Sqrrl Data, Inc
Map Reduce Register

13. LOOKING DEEPER:
REFINED BULK INGEST MODEL
Reduce Thread
Map Thread
13
Map
Setup Map Sort
Sort Reduce Output
Spill Merge
Serve
Shuffle
Time
©2014 Sqrrl Data, Inc
Parallel Latch

14. BULK INGEST MODEL PREDICTIONS
• We can constrain parts of the model by physical
throughput limitations
• Disk -> memory (100Mbps avg 7200rpm seq. read rate)
• Input reader
• Memory -> Disk (100Mbps)
• Spill, OutputWriter
• Disk -> Disk (50Mbps)
• Merge
• Network (Gigabit = 125Mbps)
• Shuffle
• And/or algorithmic limitations
• Sort, (Our) Map, (Our) Reduce, SerDe
©2014 Sqrrl Data, Inc 14

15. PERFORMANCE GOAL MODEL
Performance goals obtained through:
• Simulation of individual components
• Prediction of available resources at runtime
©2014 Sqrrl Data, Inc 15

16. INSTRUMENTATION
application version 1.3.3 SYSTEM DATA
application sha 8d17baf8 node num 1 input type arcsight
yarn.nodemanager.resource.memory-mb 43008 map num containers 20 input block size 32
yarn.scheduler.minimum-allocation-mb 2048 red num containers 20 input block count 20
yarn.scheduler.maximum-allocation-mb 43008 cores physical 12 input total 672054649
yarn.app.mapreduce.am.resource.mb 2048 cores logical 24 output map 9313303723
yarn.app.mapreduce.am.command-opts -Xmx1536m disk num 8 output map:combine input records 243419324
mapreduce.map.memory.mb 2048 disk bandwidth 100 output map:combine records out 209318830
mapreduce.map.java.opts -Xmx1638m replication 1 output map:spill 7325671992
mapreduce.reduce.memory.mb 2048 monitoring TRUE output final 573802787
mapreduce.reduce.java.opts -Xmx1638m output map:combine 7301374577
mapreduce.task.io.sort.mb 100 TIME
mapreduce.map.sort.spill.percent 0.8 map:setup avg 8 RATIOS
mapreduce.task.io.sort.factor 10 map:map avg 12 input explosion factor 13.877904
mapreduce.reduce.shuffle.parallelcopies 5 map:sort avg 12 compression intermediate 1.003327786
mapreduce.job.reduce.slowstart.completedmaps 1 map:spill avg 12 load combiner output 0.783972562
mapreduce.map.output.compress FALSE map:spill count 7 total ratio 0.786581455
mapred.map.output.compression.codec n/a map:merge avg 46
description baseline map total 290 CONSTANTS
red:shuffle avg 6 avg schema entry size (bytes) 59
red:merge avg 38
red:reduce avg 68 effective MB/sec 1.618488025
red:total avg 112
red:reducer count 20
job:total 396
©2014 Sqrrl Data, Inc 16

17. PERFORMANCE MEASUREMENT
Baseline (naive implementation)
Reduce Thread
Map Thread
Map
Setup Map Sort
Sort Reduce Output
Spill Merge
Serve
Shuffle
©2014 Sqrrl Data, Inc 17

18. PATH TO IMPROVEMENT
1. Profiling revealed much time spent serializing/
deserializing Key
2. With proper configuration, MapReduce supports
comparison of keys in serialized form
3. Rewriting Key’s serialization lead to an order-preserving
encoding, easy to compare in serialized form
4. Configure MapReduce to use native code to compare
Keys
5. Tweak map input size and spill memory for as few spills
as possible
©2014 Sqrrl Data, Inc 18

19. PERFORMANCE MEASUREMENT
Optimized sorting
• Improvements:
• Time for map-side merge went down
• Sort performance drastically improved in both
map and reduce phases
• 300% faster
©2014 Sqrrl Data, Inc 19

20. PERFORMANCE MEASUREMENT
Optimized sorting
Reduce Thread
Map Thread
Map
Setup Map Sort
Sort Reduce Output
Spill Merge
Serve
Shuffle
Insights:
• Map is slower than expected
• Output is disk bound maybe we can move more processing to Reduce
• “Reverse Amdahl’s law”
• Intermediate data inflation ratio (output/input for map) is very high
©2014 Sqrrl Data, Inc 20

21. PATH TO IMPROVEMENT
1. Profiling revealed much time spent copying data
2. Evaluation of data passed from map to reduce
revealed inefficiencies:
• Constant timestamp cost 8 bytes per key
• Repeated column names could be encoded/
compressed
• Some Key/Value pairs didn’t need to be created
until reduce
©2014 Sqrrl Data, Inc 21

22. PERFORMANCE MEASUREMENT
Optimized map code
• Improvement:
• Big speedup in map function
• Twice as fast
• Reduced intermediate inflation sped up all
steps between map and reduce
©2014 Sqrrl Data, Inc 22

23. DO TRY THIS AT HOME
Hints for Accumulo Application Optimization
With these steps, we achieved 6X speedup:
• Perform comparisons on serialized objects
• With Map/Reduce, calculate how many merge
steps are needed
• Avoid premature data inflation
• Leverage compression to shift bottlenecks
• Always consider how fast your code should run
©2014 Sqrrl Data, Inc 23

24. SOME CURRENT ACCUMULO
PERFORMANCE PROJECTS
• Optimize metadata operations
• Batch to improve throughput (ACCUMULO-2175,
ACCUMULO-2889)
• Remove from critical path where possible
• Optimize write-ahead log performance
• Maximize throughput
• Reduce flushes
• Parallelize WALs (ACCUMULO-1083)
• Avoid downtime by pre-allocating
©2014 Sqrrl Data, Inc 24

25. Securely explore your data
SQRRL IS HIRING!
QUESTIONS?
Chris McCubbin
Director of Data Science
Sqrrl Data, Inc.