bloom filter based routing for content-based publish/subscribe

Motivation Background & Overview SBSPoset & SBSTree Evaluation Summary 1 of 27 slides

Bloom Filter Based Routing for Content-BasedPublish/Subscribe

DEBS 2008

Zbigniew Jerzak and Christof Fetzer

Systems Engineering Group, Dresden University of Technology

Wednesday 2nd July, 2008

Bloom Filter Based Routing for Content-Based Publish/Subscribe Zbigniew Jerzak and Christof Fetzer


Event forwarding ([CRW01]: siena, [TK06]: forest)

100

1000

10000

100000

0 5000 10000 15000 20000 25000 30000

tim

e [

ms]

# filters

sienaforest



Subscription routing ([CRW01]: siena, [TK06]: forest)

100

1000

10000

100000

1e+06

0 5000 10000 15000 20000 25000 30000

tim

e [

ms]

# filters

sienaforest



Motivation

I Efficient End-to-End transmission delaysI More flexible edge routing approach [JF07]I A flexible communication substrate for event processingsystems



Our Contributions

I New routing structures and algorithms:I applicable for both traditional and edge routingI sbsposet – predicate storage and coverageI sbstree – management of disjunctions of predicatesconjunctions

I Content summarisation using Bloom filters [Blo70]I not imposing any limitation on the type and content of theevents/subscriptions

I sparse Bloom filters for space complexity



Publish/Subscribe Model

I Decoupled [EFGK03] communication...I ...between publishers and subscribers via routersI ...using subscriptions (conjunction of predicates)I ...and events (disjunction of predicates)I ...based on their content




I Decoupled [EFGK03] communication...I ...between publishers and subscribers via routersI ...using subscriptions (conjunction of predicates)

I

movie = ”star wars”︸︷︷︸p0()

AND price < 15︸︷︷︸p1()

I ...and events (disjunction of predicates)I ...based on their content




I Decoupled [EFGK03] communication...I ...between publishers and subscribers via routersI ...using subscriptions (conjunction of predicates)I ...and events (disjunction of predicates)

I

movie︸︷︷︸an0

= ”batman”︸︷︷︸av0

ORprice︸︷︷︸an1

= 15.5︸︷︷︸av1

I ...based on their content




I Decoupled [EFGK03] communication...I ...between publishers and subscribers via routersI ...using subscriptions (conjunction of predicates)I ...and events (disjunction of predicates)I ...based on their content



Event forwarding

An event e matches a subscription s (e - s)

∀p∈s ∃{an,av}∈e : p ({an, av}) = true (1)

A broker contains |S| subscriptions:

∀s∈S : execute Equation 1 (2)



Approach


∀p∈s ∃{an,av}∈e : p ({an, av}) = true

Evaluate predicate functions



Approach


∀p∈s ∃{an,av}∈e : p ({an, av}) = true

Calculate of disjunction of conjunctions of predicates



Overview: Subscription Routing

S

PsbsposetsbstreeR




S

PsbsposetsbstreeRp1 p2 p3




S


p1 p2 p3




S


p1 p2 p3bf1 bf2 bf3+ + +




S


p1 p2 p3bf1 bf2 bf3+ + +

bf1 bf2 bf3Bloom Filter Based Routing for Content-Based Publish/Subscribe Zbigniew Jerzak and Christof Fetzer


Overview: Event Forwarding

S

PsbsposetsbstreeR




S

PsbsposetsbstreeR an1

av1 an2av2 an3

av3




S


av1 an2av2 an3av3an1av1 an2av2 an3av3




S


av1bfi bfj bfk+ + +

an2av2 an3

av3an1av1 an2

av2 an3av3an1

av1 an2av2 an3

av3Bloom Filter Based Routing for Content-Based Publish/Subscribe Zbigniew Jerzak and Christof Fetzer



S


av1bfi bfj bfk+ + +

an2av2 an3

av3an1av1 an2

av2 an3av3an1

av1 an2av2 an3

av3 interface iinterface jBloom Filter Based Routing for Content-Based Publish/Subscribe Zbigniew Jerzak and Christof Fetzer


SBSPoset

I Stores subscriptions as:I disjunction of single predicatesI ordered by the coverage relation

I Every predicate is assigned a Bloom filterI which summarizes its contentI and that of covering predicates

I No limitations on the expressiveness of the subscription/eventlanguage



SBSPoset: subscription storage

xnull ynull≥0.7>5 <5=0{5037, 8516}{2787, 12518}{2787, 6066, 8581, 12518} {5001, 5037, 8507, 8516}Bloom Filter Based Routing for Content-Based Publish/Subscribe Zbigniew Jerzak and Christof Fetzer


SBSPoset: subscription storage x>15 y>0xnull ynull≥0.7>5 <5=0{5037, 8516}{2787, 12518}{2787, 6066, 8581, 12518} {5001, 5037, 8507, 8516}

New subscription:



SBSPoset: subscription storage x>15 y>0{5102, 8636}{6441, 10582}xnull ynull≥0.7>5 <5=0{5037, 8516}{2787, 12518}{2787, 6066, 8581, 12518} {5001, 5037, 8507, 8516}

New subscription:



SBSPoset: subscription storage y>0{5102, 8636}{6441, 10582}xnull ynull≥0.7>5>15 <5=0{5037, 8516}{2787, 12518}{2787, 6066, 8581, 12518}{6441, 10582} {5001, 5037, 8507, 8516}

New subscription: x>15



SBSPoset: subscription storage y>0{5102, 8636}{6441, 10582}xnull ynull≥0.7>5>15 <5=0{5037, 8516}{2787, 12518}{2787, 6066, 8581, 12518}{2787, 6066, 6441, 8581, 10582, 12518} {5001, 5037, 8507, 8516}

New subscription: x>15



SBSPoset: subscription storage x>15 y>0{5102, 8636}{6441, 10582}xnull ynull≥0.7>5>15 <5 >0=0{5037, 8516}{2787, 12518}{2787, 6066, 8581, 12518}{2787, 6066, 6441, 8581, 10582, 12518} {5001, 5037, 8507, 8516} {5102, 8636}

New subscription:



SBSTree

I Represents the disjunction of conjunctions of subscriptions’predicates:

I stores only Bloom filters



SBSTree: subscription storage

null5037 5001 6066 6441 27878516 8507 8581 10582 [email protected] [email protected] [email protected] [email protected] [email protected]>15 y>0{5102, 8636}{6441, 10582}New subscription:Bloom Filter Based Routing for Content-Based Publish/Subscribe Zbigniew Jerzak and Christof Fetzer


SBSTree: subscription storage

null5037 5102 5001 6066 6441 27878516 6441 8507 8581 10582 [email protected]

[email protected]@dot.com [email protected] [email protected] [email protected]>15 y>0{5102, 8636}{6441, 10582}New subscription:



SBSPoset: event forwarding x=7 y=-2xnull ynull≥0.7>5>15 <5 >0=0{5037, 8516}{2787, 12518}{2787, 6066, 8581, 12518}{2787, 6066, 6441, 8581, 10582, 12518} {5001, 5037, 8507, 8516} {5102, 8636}

Event:



SBSPoset: event forwarding y=-2xnull ynull≥0.7>5>15 <5 >0=0{5037, 8516}{2787, 12518}{2787, 6066, 8581, 12518}{2787, 6066, 6441, 8581, 10582, 12518} {5001, 5037, 8507, 8516} {5102, 8636}

Event: x=7



SBSPoset: event forwarding y=-2xnull ynull≥0.7>5>15 <5 >0=0{5037, 8516}{2787, 12518}{2787, 6066, 8581, 12518}{2787, 6066, 6441, 8581, 10582, 12518} {5001, 5037, 8507, 8516} {5102, 8636}

Event: x=7{2787, 6066, 8581, 12518}



SBSPoset: event forwarding

xnull ynull≥0.7>5>15 <5 >0=0{5037, 8516}{2787, 12518}{2787, 6066, 8581, 12518}{2787, 6066, 6441, 8581, 10582, 12518} {5001, 5037, 8507, 8516} {5102, 8636}

Event: x=7{2787, 6066, 8581, 12518}y=-2



SBSPoset: event forwarding

xnull ynull≥0.7>5>15 <5 >0=0{5037, 8516}{2787, 12518}{2787, 6066, 8581, 12518}{2787, 6066, 6441, 8581, 10582, 12518} {5001, 5037, 8507, 8516} {5102, 8636}

Event: x=7{2787, 5037, 6066, 8516, 8581, 12518}y=-2



SBSTree: event forwarding null5037 5102 5001 6066 6441 27878516 6441 8507 8581 10582 [email protected]

[email protected]@dot.com [email protected] [email protected] [email protected]: x=7{2787, 5037, 6066, 8516, 8581, 12518}y=-2



Evaluation Environment

I SIENA 1.5.4I http://www.inf.unisi.ch/carzaniga/siena/

I Apache MinaI Stochastic Simulation in Java (SSJ) libraryI BloomFilter


http://www.inf.unisi.ch/carzaniga/siena/



I SIENA 1.5.4I Apache Mina

I http://mina.apache.org/I ver. 2.0.0-M1 as of 24th Jan 2008

I Stochastic Simulation in Java (SSJ) libraryI BloomFilter


http://mina.apache.org/



I SIENA 1.5.4I Apache MinaI Stochastic Simulation in Java (SSJ) library

I umontreal.iro.lecuyer.randvar.*I umontreal.iro.lecuyer.rng.*

I BloomFilter




I SIENA 1.5.4I Apache MinaI Stochastic Simulation in Java (SSJ) libraryI BloomFilter

I http://wwwse.inf.tu-dresden.de/xsienaI Based on code by: Hongbin Liu and Arash Partow


http://wwwse.inf.tu-dresden.de/xsiena


Event forwarding

10

100

1000

10000

100000

0 5000 10000 15000 20000 25000 30000

tim

e [m

s]

# filters

sienaforest

sbsposetsbstree



Subscription routing

10

100

1000

10000

100000

1e+06

0 5000 10000 15000 20000 25000 30000

tim

e [m

s]

# filters

sienaforest

sbsposetsbstree



TCP/IP Latency

0

200

400

600

800

1000

1200

1400

1600

0 1000 2000 3000 4000 5000 6000 7000

num

ber

of

even

ts

latency [µs]

1 broker (local)2 brokers (local)3 brokers (local)

1 broker (distributed)2 brokers (distributed)3 brokers (distributed)



TCP/IP Throughput

0

2

4

6

8

10

12

14

16

18

1 2 3 4 5 6 7 8 9 10 0

2

4

6

8

10

12

K e

vents

/ s

econd

MB

/s

K events sent

local (1 broker)local (2 brokers)local (3 brokers)

distributed (1 broker)distributed (2 brokers)distributed (3 brokers)



Local Throughput

0K

100K

200K

300K

400K

500K

10K 20K 30K 40K 50K 60K 70K 80K 90K 100K

events

/second

# events

unfrm distr norm distr Pareto distr

1K

10K

100K

deli

evere

d e

vents



False Positives

10-6

10-5

10-4

10-3

10-2

100K 200K 300K 400K 500K 600K 700K

1000 1500 2000 2500 3000 3500 4000 4500 5000

% f

alse

posi

tives

# delivered events

# subscriptions

% false pos, log2m=12, k=4% false pos, log2m=13, k=4% false pos, log2m=13, k=5% false pos, log2m=13, k=6



Bloom Filter vs Sparse Bloom Filter

103

104

105

106

0 20000 40000 60000 80000 100000 120000

mem

ory

consu

mpti

on [

KB

]

# elements inserted

Bloom filter Sparse Bloom filter



Summary

I Low latency, high throughputI Event forwarding based on Bloom filtersI No restrictions on the subscription languageI Decoupled management of predicates and their conjunctions



Thank You!http://wwwse.inf.tu-dresden.de/xsiena/


http://wwwse.inf.tu-dresden.de/xsiena/


References

Burton H. Bloom.

Space/time trade-offs in hash coding with allowable errors.Communications of the ACM, 13(7):422–426, 1970.

Antonio Carzaniga, David S. Rosenblum, and Alexander L. Wolf.

Design and evaluation of a wide-area event notification service.ACM Trans. Comput. Syst., 19(3):332–383, 2001.

Patrick Th. Eugster, Pascal A. Felber, Rachid Guerraoui, and Anne-Marie Kermarrec.

The many faces of publish/subscribe.ACM Comput. Surv., 35(2):114–131, 2003.

Zbigniew Jerzak and Christof Fetzer.

Prefix forwarding for publish/subscribe.In DEBS ’07: Proceedings of the 2007 Inaugural International Conference on Distributed Event-BasedSystems, pages 238–249, Toronto, Ontario, Canada, June 2007. ACM Press.

Sasu Tarkoma and Jaakko Kangasharju.

Optimizing content-based routers: posets and forests.Distributed Computing, 19(1):62–77, September 2006.


bloom filter based routing for content-based publish/subscribe

Technology

content bloom filter

slides bloom filter

slides event

space complexity bloom

filtersbloom filter

bloom lters blo70not

event e

s subscriptions