OrientDB vs Neo4j - Comparison of query/speed/functionality

OrientDB vs Neo4j
Comparisons (querys and functionality)
Curtis Mosters
@02.12.2014

Content
• Schema
• Indexes
• Comparison
• Query/Speed
• Functionality
• Results
2OrientDB vs Neo4j - Comparison

Prototype Comparison
Schema
ApplnPerson
WROTE
Abstract
HAS_ABSTRACT
ID:INTEGER
name:String
ID:INTEGER
title:String
ID:INTEGER
abstract:String

Indexes
• Appln.title
• LUCENE FULLTEXT
• Appln.ID
• SBTREE UNIQUE (in Neo4j the usual INDEX)
• Person.title
• LUCENE FULLTEXT
• Person.ID
• SBTREE UNIQUE (in Neo4j the usual INDEX)

ComparisonPrototype
Querys and used systems
• comparing the speed of both on typical requests
• Linux 64-bit (same instance on AWS)
• OrientDB v.2.0M2
• Neo4j v.2.1.5
• Speed tests are done in the same order the slides/rows are
• One database per instance  2 instances
• Servers are idling and just OrientDB/Neo4j running
• Querys are tested by hand on the command line (not in the studio)
• Querys always having the same results on both databases
• Times are always given in milliseconds (ms) if not specified
• Both databases using the StandardAnalyzer from Lucene
• Cache cleared after querys

ComparisonPrototype
System cache notes
• OrientDB is always clearing the cache when restarted
• Neo4j does not clear the cache
• So in the Neo4j column I in some cases tested with cleared system cache and sometimes
without
• If there is just one column on Neo4j it is „No System cache cleared“

Comparison (Query/Speed)
OrientDB vs Neo4j - Comparison 7

ComparisonPrototype
Import
OrientDB
• Official supported methods
• OrientDB-ETL/JDBC
• Java API
• Clean Java code
• ETL tool is performant but at last tests having
issues with edge creation
• Not using Multi-Threading
• Not using Mapping
Neo4j
• Official supported methods
• LOAD CSV command
• Java API
• Groovy
• Batch-Importer
• Talend
• No really „easy“ way but Java is the fastest and
most reliable way
• Using Multi-Threading and Mapping
OrientDB vs Neo4j - Comparison8
~300mio lines {APPLNs,TITLEs,PERSONs} with edges and indexes
25 hours 19 hours

ComparisonPrototype
Startup/Shutdown speed
OrientDB
• Nearly always the same time when starting or
shutting down the server
• 2 sec – 10 sec
Neo4j
• Different times when starting and especially by
shutting down the server when task is still
running
• 3 sec – 3 min (no infos)
Good for testing and later reliability

ComparisonPrototype
Query #1
OrientDB Neo4j
Checking Single ID lookup
? SELECT FROM Appln WHERE ID=? MATCH (a:Appln)WHERE a.ID=? RETURN a
1412 27 71 939
763773 9 30 44
234526 15 26 43
858584 10 25 44
536367 11 25 43
2323 17 18 31
5267 1 15 24
73573 14 29 35
585985 10 25 34
797977 10 26 35
Average 12,4 (10 of 10) 29 (0 of 10)
No system cache cleared System cache cleared

ComparisonPrototype
Query #2
OrientDB Neo4j
Checking Fulltext Lucene Lookup
?
Note on Neo4j:
more than one word needs to
be put in a new property
statement, e.g. instead of
'title:super efficient'
we take 'title:super OR
title:efficient'
SELECT FROM (SELECT title,ID FROM ApplnWHERE title
LUCENE "?" ORDER BY ID) LIMIT 10
START n=node:titles('title:?') RETURN n.title,n.IDORDER BY n.ID
LIMIT 10
solar 10172 801 137088
panel 263698 121494 161215
druck 25582 9679 11290
machine 1146339 297645 357818
cell 253565 55397 26298
automatic vehicle 961054 131772 163794
super efficient 53380 8432 8707
motor 398803 79527 46687
airplane 14066 892 390
windshield 8969 1004 536
Average 313 sec (5,2 min) (0 of 10) 70 sec (10 of 10)

ComparisonPrototype
Query #3.1
OrientDB Neo4j
Checking Fulltext Lucene Lookup Overall Count on 1 indices
?
Note on Neo4j:
title:efficient'
SELECT $totalHits
FROMAppln
WHERE title LUCENE "?" LIMIT 1
START n=node:titles("title:?")
RETURN count(*)
solar 4611 215263
panel 3318 77442
druck 2890 12503
machine 1846 198479
cell 2351 34685
automatic vehicle 1063 49283
super efficient 984 4054
motor 465 47085
airplane 1172 429
windshield 62 585
Average 9 of 10 1 of 10

ComparisonPrototype
Query #3.2
OrientDB Neo4j
Checking Fulltext Lucene Lookup Overall Count on 2 indices
?
Note on Neo4j:
title:efficient'
SELECT $totalHits
FROMAppln
WHERE [title,abstract] LUCENE "?" LIMIT 1
START n=node:titles ('title:?')
MATCH (n)-[:HAS_ABSTRACT]->(a)WHERE a.abstract =~ ".*?.*"
RETURN count(*)
solar 227234
panel
druck
machine
cell
automatic vehicle
super efficient
motor
airplane
windshield
Average

ComparisonPrototype
Query #4
OrientDB Neo4j
Internal ID function node lookup
?
OrientDB
?
Neo4j
SELECT title FROM #11:? / SELECT name FROM #12:? START n=node(?) RETURN n.title / START n=node(?) RETURN
n.name
11:0 0 1 10 816
11:141 141 1 13 27
11:26526 26526 3 13 28
11:2526 2526 2 12 27
11:6262 6262 1 12 28
12:0 76594275 1 11 25
12:515 76594790 2 14 23
12:4115 76598390 3 14 25
12:52627 76646902 2 13 26
12:47484 76641759 1 13 25
Average 2 (10 of 10) 13 (0 of 10)

ComparisonPrototype
Query #5
OrientDB Neo4j
Count Applns of a specific Person
?
OrientDB
?
Neo4j
SELECT out(WROTE).size()
FROM #?
START p=node(?)
MATCH (p)-[:WROTE]->(a)
RETURN count(*)
12:0 76594275 8 81 980
12:1 76594276 1 18 42
12:2 76594277 1 20 41
12:3 76594278 1 18 38
12:4 76594279 1 17 39
12:5 76594280 1 23 41
12:6 76594281 1 21 37
12:7 76594282 1 17 43
12:8 76594283 1 18 45
12:9 76594284 1 17 41
Average 1 (10 of 10) 25 (0 of 10)

ComparisonPrototype
Query #6
OrientDB Neo4j
Searching for 3 Applns of one specific Person
?
OrientDB
?
Neo4j
select out.@class as sourceClass,out.@rid as source ,out.name
as sourceName,in.@class as targetClass,in.@rid as target,in.ID
as targetID ,in.nrEpodoc as targetName from (select
expand(outE('WROTE')) from #?) order by targetID ASC limit 3
START p=node(?)
RETURN labels(p) as sourceClass, id(p) as source, p.name as
sourceName, labels(a) as targetClass, id(a) as target, a.nrEpodoc
as targetNameORDER BY a.ID ASC LIMIT 3
12:0 76594275 1051 107 212
12:1 76594276 3 39 77
12:2 76594277 2 40 68
12:3 76594278 2 38 60
12:4 76594279 3 41 58
12:5 76594280 53 59 55
12:6 76594281 56 53 59
12:7 76594282 7 38 56
12:8 76594283 5 38 62
12:9 76594284 2 33 66
Average 118 (8 of 10) 49 (2 of 10)

ComparisonPrototype
Query #7
OrientDB Neo4j
Searching for Appln.title and Appln.abstract
return Person.name matching both
?
Title
SELECT FROM (SELECT title,abstract,ID from Appln
where [title,abstract] LUCENE "?" ORDER BY ID) LIMIT 3
START p=node:titles('title:?')
MATCH (p)-[:HAS_ABSTRACT]->(a) WHERE a.abstract
=~ ".*?.*"
RETURN p.title,a.abstract,a.ID ORDER BY a.ID LIMIT 3
panel 1733261 424789
Average

ComparisonPrototype
Query #7
OrientDB Neo4j
Searching a Person.name + searching on Appln.title for Appln of that specific Person
return Person.name matching both
?
Title
START p=node:people('name:?')
MATCH (p)-[:WROTE]->(a) WHERE a.title =~ ".*?.*"
RETURN p.name,a.title,a.IDORDER BY a.ID LIMIT 3
machine 99538
Average

ComparisonPrototype
Query #8
OrientDB Neo4j
Searching for an Abstract of an Appln
?
Note on Neo4j:
title:efficient'
select @rid,abstract,ID as titleID,in(HAS_ABSTRACT).title as
title,in(HAS_ABSTRACT).ID as AbstrID fromAbstract where
abstract LUCENE "method" LIMIT 3
START n=node:abstracts("abstract:method")
WITH n limit 3
MATCH (x:Appln)-[:HAS_ABSTRACT]->(n)
RETURN n.ID,x.ID
solar
panel
druck
machine
cell
automatic vehicle
super efficient
motor
airplane
windshield
Average

ComparisonPrototype
Query #9
OrientDB Neo4j
Counting the Applns of Person.names containing a specific name
? SELECT sum(out(WROTE).size())
FROM Person
WHERE name LUCENE "?" LIMIT -1
START p=node:people('name:?')
RETURN count(a)
bosch 7475 3771
intel 13261 7461
siemens 19302 16297
audi 3888 1844
volkswagen 2872 1298
toyota 23223 13561
sony 16520 11449
panasonic 6314 2287
microsoft 2849 1313
apple 3127 1088
Average 0 of 10 10 of 10

Comparison (Functionality)

ComparisonPrototype
Database Overview
OrientDB
• Schema, naming policies, overall records,
cluster infos and many more infos
• Whole page in 0,1 sec
Neo4j
• No schema infos except naming policies
• Counting single label nodes takes ~10 min
Easy and fast way to check state of the database Neo4j‘s supported way to get infos on all
labels in one query just gives a Heap Error
(maybe too much data?)

ComparisonPrototype
Graph Explorer
OrientDB
• Good overview, straightforward and fast
• Nodes can be edited, edges added
• Never-ending-graph like
Neo4j
• Showing nodes/edges and when being clicked
some infos about
• No other features, not even zooming or
dragging all elements
Good for checking graph issues as near as possible to the database
v.2 only!

ComparisonPrototype
Result view
OrientDB
• Great overview and paging possible to lower
showup and query speed
• If you miss setting a „LIMIT“ it‘s set for you!
• Using new GraphTab for visual things (v.2!)
Neo4j
• Graph andTable view
• Miss setting a LIMIT? Go smoking 
• Graph just able to see up to 10 nodes
• Table view endless scrolling
Getting an overview is quite important to check specific query issues

ComparisonPrototype
Function integration
OrientDB
• Good overview and management
• Integrated in the Studio
• No restart needed
• Functions can even be copied to another db
Neo4j
• Server plugins [1]
• Needs to be written in Java and inherited from
ServerPlugin class
• No overview
• Not fail-save
• No easy change/access
• Requires Server restart
• Many lines for simple things
Needed for exchange information with the prototype

ComparisonPrototype
Query style
OrientDB
• Simple querys really short
• Hard to write querys when they are getting
complex
• Bad overview and using variable names not
intuitive
Neo4j
• Simple querys really long due to needed
cypher statements
• Easy to write also complex querys
• Using variables name is very intuivite and
always keeping up the overview
Useful for result checking and testings

ComparisonPrototype
Lucene Index
OrientDB
• Still a „new“ addon
• Prior v.2 plugin needed
• With v.2 integreated in OrientDB
• Use it as if you set an usual index
• Index can easily be changed at any time
• Analyzer can be easily changed
Neo4j
• Neo4j does not always use Lucene as indexer
• Needs to be set before importing data
• Works together via node_auto_index
configuration
• Changing index or set index to Lucene after the
import is not viable in terms of time aspects
• Analyzer is not easy to change
Important for full text search the new graph tab builds up

ComparisonPrototype
Security
OrientDB
• Different security levels (like in MySQL)
Neo4j
• None
Good for integrating more databases and setting access levels

ComparisonPrototype
Disc usage
OrientDB
• Db size = 120 GB
• Classes in different files
• Classes can also be easily deleted by external
deletion
Neo4j
• Db size = 40 GB
• Nodes, properties and relations in separate
files
• Specific data can only be deleted by Neo4j
commands
Good for testing and later reliability

ComparisonPrototype
Future Perspective
OrientDB
• OrientDB still „new“ on the market, many
features still coming
• Still much place for improvements
• Brings the possibility to replace MySQL
Neo4j
• Neo4j „oldest“ Graph database and nearly any
feature in there
• Algorithms already improved as best as
possible
• No possiblity to replace a current system, just
an extension for using graphs
To see ahead of the current state

ComparisonPrototype
Costs
OrientDB
• Good support for free available
• Commercial support much cheaper than Neo4j
• EnterpriseVersion available with good
monitoring features
Neo4j
• Commercial support needed to setup a well
defined database
• Features like clustering only available when
paying (e.g. important for our where clause)
Important for startups

ComparisonPrototype
Support / Production speed / Own Ideas
OrientDB
• Good support via
• E-Mail
• Google Group (anyone from the team helping)
• Gitter
• Github
• Every 2-3 weeks new release
• Own Issues answered in 1-2 day
• Own ideas are discussed, every day 30-40
comments in Github
Neo4j
• Poor support for the most popular graph db
• Google Group only semi-active community
• Just one member from Neo4j helping there
• Every 1-2 month new release
• Own issues answered ~1 week
• Own ideas are mainly ignored, every day 20-30
comments in Github
Important for later issue solvings

Results (Speed)
Measure OrientDB Neo4j
Import no use of MT/mapping full use of MT/mapping
Startup/Shutdown Speed x -
Query #1 Checking Single ID lookup x -
Query #2 Checking Fulltext Lucene Lookup - x
Query #3.1 Checking Fulltext Lucene Lookup Overall Count on 1 indices x -
Query #3.2 Checking Fulltext Lucene Lookup Overall Count on 2 indices - -
Query #4 Internal ID function node lookup x -
Query #5 Count Applns of a specific Person x -
Query #6 Searching for 3 Applns of one specific Person single bolter making poor average value always quite same speed
Query #7 Searching a Person.name + searching on Appln.title for Appln - -
Query #8 Searching for an Abstract of an Appln - -
Query #9 Counting the Applns of Person.names containing a specific name - x
Results 4 3

Results (Misc)
Measure OrientDB Neo4j
Database Overview x
Graph Explorer x
Result View x
Function Integreation x
Query style x
Lucene Index x
Security x
Disc Usage every class in single file using less disk space
Future Perspective x
Costs x
Support / Production Speed / Own ideas x
Results 9 1

Results
• OrientDB working on fixing the very slow querys
• OrientDB has inconsistent query speed somtimes (super high and super low)
• OrientDB Studio is on a really next level
• Neo4j Studio nearly useless compared to OrientDB‘s

Supporters
• I want to give a special thanks to Michael Hunger, without him the Neo4j
import would still have trouble
• I also want to thank Enrico Risa for his help and fast implementation of
Lucene improvements
• Keep up the great work!

Links
• [1] http://docs.neo4j.org/chunked/stable/server-plugins.html
• [2] http://docs.neo4j.org/refcard/2.0/

OrientDB vs Neo4j - Comparison of query/speed/functionality

More Related Content

What's hot

Similar to OrientDB vs Neo4j - Comparison of query/speed/functionality

Recently uploaded

OrientDB vs Neo4j - Comparison of query/speed/functionality