i need the data and i need it 1ms ago!

Angelo Corsaro, Ph.D.Chief Technology Officer

OMG DDS SIG [email protected]

Delivering Performance, Openness, and Freedom

OpenSplice DDS

I Need Market Data and I need it 1ms Ago!

mailto:[email protected]


D e l i v e r i n g P e r f o r m a n c e , O p e n n e s s , a n d F r e e d o m

OpenSplice DDS

Challenges

© 2009, PrismTech. All Rights Reserved

Financial FirmsTrends

‣ Direct connection to exchanges is increasingly chosen over data aggregators to reduce trading latency

‣ Market data is growing steadily

‣ Milliseconds speedup over competitors can be worth millions of dollars

Challenges

‣ Process several hundreds-of-thousands of messages, quickly approaching millions messages, per second

‣ Minimize latency and jitter, while maintaining the system stability and reliability

Deployed middleware solutions were not designed to address these challenges!


Exchanges

Trends

‣ More and more firms connecting directly

‣ Increasing number of sources providing the same quotes

‣ Competitive advantage obtained by reducing latency on data feeds

Challenges

‣ Ensuring fairness of data distribution

‣ Minimizing latency

‣ Scaling with the number of destinations

Deployed middleware solutions were not designed to address these challenges!


OPRA PredictionsTrends

‣ One order of magnitude growth on market data from 2005 to 2008

‣ OPRA message rates predicted to reach ~2M msgs/sec by early 2009

‣ Traffic bursts predicted to further increase by 20%

Challenges

‣ Can your messaging infrastructure keep-up with this trend?

0

500000

1000000

1500000

2000000

88000

190000

359000

701000

1807000

Jan 13, 2005

May 16, 2006

Jan 9, 2007Jan 8, 2008

Jan 6, 2009

Msgs/sec

OPRA Market Data Rates Forecasts


Trading with Pareto‣ Usually 20% of the trades

provides 80% of the gains

‣ Under overload condition how can you ensure you keep trading at least 20% of your target?

‣ How can you ensure that you keep trading the the right 20% of data?

Vilfredo Pareto, 1848-1923

Average

Time

Msgs/sec

System Message Rate

Incoming Message Rate


OpenSplice DDS

The Data Distribution Service for Real-Time Systems (DDS)


Addressing Data Distribution Challenges

The OMG DDS Standard‣ Introduced in 2004 to address the Data

Distribution challenges faced by a wide class of Defense and Aerospace Applications

‣ Key requirement for the standard were its ability to deliver very high performance while seamlessly scaling from embedded to ultra-large-scale deployments

‣ Today recommended by key administration worldwide and widely adopted across several different application domains, such as, Automated Trading, Simulations, SCADA, Telemetry, etc.

DDS is standard designed to address the data-distribution challenges across a wide class of Defense and Aerospace Applications


The OMG Data Distribution Service (DDS)

DDS v1.2 API Standard‣ Language Independent, OS and HW architecture

independent

‣ DCPS. Standard API for Data-Centric, Topic-Based, Real-Time Publish/Subscribe

‣ DLRL. Standard API for creating Object Views out of collection of Topics

DDSI/RTPS v2.1 Wire Protocol Standard‣ Standard wire protocol allowing interoperability

between different implementations of the DDS standard

‣ Interoperability demonstrated among key DDS vendors in March 2009

Object/Relational Mapping

Ownership DurabilityContent

Subscription

Minimum Profile

Data Centric Publish/Subscribe (DCPS)

Data Local Reconstruction Layer (DLRL)

DDS Interoperability Wire Protocol

Application

UDP/IP

Real-Time Publish/Subscribe Protocol


Scope of Standardization



Subscription

Minimum Profile




Application

UDP/IP


DD

SI

v2

.1D

DS

v1

.2

Interoperability Wire Protocol

(TCP, SCTP, UDP)/IP

Advanced Message Queuing Protocol

AM

QP

v0

.10

API (?)

Application

AMQP OMG DDS





Subscription

Minimum Profile




Application

UDP/IP


DD

SI

v2

.1D

DS

v1

.2


(TCP, SCTP, UDP)/IP


AM

QP

v0

.10

API (?)

Application

Wire Protocol Standard

AMQP OMG DDS





Subscription

Minimum Profile




Application

UDP/IP


DD

SI

v2

.1D

DS

v1

.2


(TCP, SCTP, UDP)/IP


AM

QP

v0

.10

API (?)

Application


Pub/Sub Standard API

AMQP OMG DDS





Subscription

Minimum Profile




Application

UDP/IP


DD

SI

v2

.1D

DS

v1

.2


(TCP, SCTP, UDP)/IP


AM

QP

v0

.10

API (?)

Application


Pub/Sub Standard API

Object Relational Mapping Standard API

AMQP OMG DDS


Architectural Style

Publisher

SubscriberPublisher

Publisher

Subscriber

Subscriber

Broker

OMG DDS

Publisher

SubscriberPublisher

Publisher

Subscriber

Subscriber

Broker

Broker

Broker

Broker

Broker

AMQP


Key Differences in Summary

‣ Peer-to-Peer Communication Architecture

‣ Wire Protocol + API Standard‣ Interoperability across DDS implementations

‣ Application Portability across DDS Implementation

‣ Content-Based Filtering/Quering

‣ Designed bottom-up for maximum performance and determinism

DDS‣ Brokered Communication

Architecture

AMQP

‣ Wire Protocol Standard‣ Interoperability across AMQP

implementations

‣ Header Based Routing

‣ Not geared for High-Performance


OpenSplice DDS

DDS and Open Source

Enterprise Ed.

Pro

prie

tary

Info

rmat

ion

- D

istri

butio

n w

ithou

t Exp

ress

ed W

ritte

n P

erm

issi

on is

Pro

hibi

ted.


OpenSplice DDS

‣ Product reorganized into Editions‣ Community Edition

‣ Compact Edition

‣ Professional Edition

‣ Enterprise Edition

‣ Product Editions provide a a growing set of functionality to address the needs of increasingly more sophisticated users

‣ The Community Edition is Open Source

‣ Compact, Enterprise and Professional Edition are available only through Commercial Subscriptions

Enterprise Ed.

Community Ed.

Compact Ed.

Professional Ed.

Pro

prie

tary

Info

rmat

ion

- D

istri

butio

n w

ithou

t Exp

ress

ed W

ritte

n P

erm

issi

on is

Pro

hibi

ted.


Community EditionFeatures

‣ OMG DDS v1.2 DCPS‣Minimum Profile

‣ Content Subscription Profile

‣ Durability Profile

‣ Ownership Profile

‣ Networking‣ DDSI v2.1 Implementation

‣ Real-Time Networking Implementation


Subscription

Minimum Profile

Real-Time Pub/Sub (DCPS)

Networking Technology

Application

UDP/IP

Interoperable Wire Protocol (DDSI)

DD

SI v

2.1

DD

S v

1.2

Real-Time Networking

Open Source LGPL Licensed Edition -- The Best Way to Get Started with DDS

Licensing

‣ LGPLv3

Pro

prie

tary

Info

rmat

ion

- D

istri

butio

n w

ithou

t Exp

ress

ed W

ritte

n P

erm

issi

on is

Pro

hibi

ted.


Performance on Commodity HW

!"#$%&#'

()(!%(*'

&")&%+)'

&!(,%!)'

),,%$#'

!,*%))'(("%!"'

&&)%*$' $"%"$' ,+%($',($%!+'

,*+%&)'

$&"%,,'

*!#%,&'

)$,%$*'

)"#%*)'

"!+%++'

"*(%)+' ")(%&!' ""&%,&'

,++%++'

!++%++'

$++%++'

#++%++'

*++%++'

)++%++'

"++%++'

&+++%++'

&&++%++'

(+%++'

&+(+%++'

(+(+%++'

,+(+%++'

!+(+%++'

$+(+%++'

#+(+%++'

)' &#' ,(' #!' &()' ($#' $&(' &+(!' (+!)' !+"#'

!"#$%

&'$($)$*+%

!*$$,(*%-./*%

-./0/1/23' 456/'

Test Scenario

‣ Single Threaded Application (multi-threaded networking service)

‣ 8192 bit message batches

Throughput Latency

Inter-Node Latency

‣ 60 usec

Inter-Core Read-Latency

‣ 2 usec

Inter-Core Latency

‣ <10 usec

HW:

‣ Dell blade-server

‣ Dual-core, Dual-CPU, AMD Opteron 2.4 GhzOS

‣ Linux 2.6.21-1.3194.fc7Network

‣ Gigabit Ethernet cards

‣ Dell PowerConnect 5324 switch

Pro

prie

tary

Info

rmat

ion

- D

istri

butio

n w

ithou

t Exp

ress

ed W

ritte

n P

erm

issi

on is

Pro

hibi

ted.


OpenSplice DDS AdoptionDefense

‣ Combat Management Systems

‣ Tactical Systems

‣ Drones (land/air)

‣ Communication Systems

‣ Radar Processing

‣ Submarine Systems

‣ Future Combat Systems

SCADA/Utilities

‣ Industrial Automation

‣ Power Grids (Smart Grids)

‣ Complex Telemetry (e.g. NASA Rocket Launch System)

Pro

prie

tary

Info

rmat

ion

- D

istri

butio

n w

ithou

t Exp

ress

ed W

ritte

n P

erm

issi

on is

Pro

hibi

ted.


OpenSplice DDS Adoption

Transportation

‣ Air Traffic Management/Control

‣ Metropolitan Traffic Management

‣ Underground Metropolitan Transportation

Financial Services

‣ Automated Trading

‣ Risk Management

‣ Option Trading


Some OpenSplice DDS Users

!"#$%&'()('*+'##,-+.+/($$,012"0

!"(-/'0&3+4"25&0+-2",6+"78+29:;<=7>?

@A6BC+/DEF=7>>;>

4"25&0

'0,G+HII+JCKA+B@+CA+LI+L6

4DM+G+HII+JCKA+B@+CA+LI+LL

10NG!""#$%%&&&'#()*+",-!'. .-/+. .

Marcoussis, le 28/08/09

Expertise pour la RATP

réalisée par Bernard MAUDRY les 29 et 29 juillet 2009

La RATP va utiliser OpenSplice/DDS pour remplacer son système AGS de gestion des équipements de

station.

Dans ce cadre, elle a demandé à PrismTech d'effectuer une expertise de son ébauche de conception du

nouveau système et de fournir des informations complémentaires et des conseils afin d'exploiter au

mieux les possibilités d'OpenSplice/DDS.

La première journée a été consacrée à la présentation de la conception préliminaire, à sa correction et

aux suggestions concernant son évolution. La conception initiale a été fortement impactée et les

modifications nécessaires impliqueront probablement une nouvelle revue afin de valider les derniers

détails.

La deuxième journée a principalement été dédiée à la réponse aux questions de comportement

d'OpenSplice/DDS dans les situations les plus délicates que rencontrera le nouveau système, en tenant

compte de l'architecture choisie.

Cette architecture comprend les éléments suivants:

! le système est structuré en «lignes» comprenant un centre de contrôle de ligne (niveau 2) et des

stations (niveau 1). Une ligne de métro ou de RER peut être décomposée en plusieurs «lignes»,

auquel cas un des centre de contrôle peut être amené à supplanter un ou plusieurs des autres de

cette ligne.

! Des centres de niveau 3 peuvent intervenir sur n'importe quelle ligne

? Le centre de maintenance (Toscane)

? Le centre de gestion de la ventilation

? Des postes d'intervention (action à partir de n'importe quel endroit du réseau)

! Une adresse IP classe A est associée à chaque ligne

! Une station comprend une noeud de gestion (redondé) chargé de la gestion des équipements de

la station et un poste opérateur

0123.14..-5#) "56 .7,.89.:;;<=>[email protected](C.D.EFG.H=H.9GG.>>>E8.I.+5)*/J .+K(, . $ [email protected]<.2ARM0OA2AQ.MS.ASR31SQ.$!ST.G==9E=H'. .2ARM0OA2AQ.TUUM@A.$ [email protected]<.HO P.1WASVA.DV0MSA00.L12X< .OA1N.W133AY<.R1OA0PA1Q< .SA::.>SR


OpenSplice DDS

Understanding DDS


As Simple as it Gets‣ DDS is based around the concept

of a fully distributed Global Data Space (GDS)

Global Data Space

DDS

Brokers




‣ Publishers and Subscribers can join and leave the GDS at any time

Global Data Space

DDS

Brokers





Global Data Space

DDS

Subscriber

Subscriber

Subscriber

Publisher

Publisher

Publisher

Brokers





‣ Publishers and Subscribers express their intent to produce/consume specific type of data, e.g., Topics

Global Data Space

DDS

Subscriber

Subscriber

Subscriber

Publisher

Publisher

Publisher

Brokers





‣ Publishers and Subscribers express their intent to produce/consume specific type of data, e.g., Topics

‣ Data flows from Publisher to Subscribers

Global Data Space

DDS

Subscriber

Subscriber

Subscriber

Publisher

Publisher

Publisher

Brokers


OpenSplice DDS

Defining Data


DDS Topics

Topic

‣ Unit of information exchanged between Publisher and Subscribers.

‣ An association between a unique name, a type and a QoS setting

struct ShapeType { long x; long y; long shapesize; string color;};#pragma keylist ShapeType color

{ShapeType}

{Circle, Square, Triangle}

{...}

Topic Type.

‣ Type describing the data associated with one or more Topics

‣ A Topic type can have a key represented by an arbitrary number of attributes

‣ Expressed in IDL (or XML)


DDS Topics

Circle


Topic QoS


DDS Topics

Triangle


Topic QoS


DDS Topic Instances and Samples

Topic Instances

‣ Each key value identifies a unique Topic Instance,

‣ Topic’s instance lifetime can be explicitly managed in DDS

Topic Samples

‣ The values assumed by a Topic Instance over time are referred as Instance Sample



Topic/Instances/Samples Recap.Topics Instances

Samples

timeti tj tnow


Content Filtering

‣DDS allows to specify content-filtered Topics for which a subset of SQL92 is used to express the filter condition

‣Content filters can be applied on the entire content of the Topic Type

‣Content filters are applied by DDS each time a new sample is produced/delivered

X0 X1

Y0

Y1

X0 <= X <= X1

Y0 <= Y <= Y1


Local Queries‣ Subscribed Topics can be seen locally as “Tables”

‣ A subset of SQL92 can be used for performing queries on multiple topics as well as natural joins

Circle Topic

SELECT * FROM ShapeType s WHERE s.x > 25 AND s.y < 55

color x y shapesizeyellow 30 25 50

color x y shapesizeredblue

yellow

57 62 5090 85 5030 25 50

‣ Queries are performed under user control and provide a result that depends on the current snapshot of the system, e.g., samples currently available

X0

Y0Y0

Y1

Y0 <= Y <= Y1


OpenSplice DDS

Anatomy of a DDS App.


Anatomy of a DDS Application

1 21 62

2 20 61

3 25 70 25

1 22 62 1 23 63

2 19 60

3 25 71 3 25 74 3 26 77

Samples

Topic

Instances1 21 62

2 20 61

3 25 70 25

1 22 62 1 23 63

2 19 60

3 25 71 3 25 74 3 26 77

Samples

Topic

InstancesSamples

Topic (e.g. Circle)

InstancesDataReaderDataReaderDataReaderDataReaderDataReaderDataReaderDataWriterDataWriterDataWriterDataWriter

blue 21 62 80


blue 23 66 80 blue 27 70 80

orange 21 62 80

green 97 75 50 green 94 71 50

orange 21 62 80



Arrows show structural relationships, not data-flows Partition

1 21 62

2 20 61

3 25 70 25

1 22 62 1 23 63

2 19 60

3 25 71 3 25 74 3 26 77

Samples

Topic

Instances1 21 62

2 20 61

3 25 70 25

1 22 62 1 23 63

2 19 60

3 25 71 3 25 74 3 26 77

Samples

Topic

InstancesSamples

Topic (e.g. Circle)


blue 21 62 80


blue 23 66 80 blue 27 70 80

orange 21 62 80

green 97 75 50 green 94 71 50

orange 21 62 80



Partition

Domain

Domain Participant

Arrows show structural relationships, not data-flows Partition

1 21 62

2 20 61

3 25 70 25

1 22 62 1 23 63

2 19 60

3 25 71 3 25 74 3 26 77

Samples

Topic

Instances1 21 62

2 20 61

3 25 70 25

1 22 62 1 23 63

2 19 60

3 25 71 3 25 74 3 26 77

Samples

Topic

InstancesSamples

Topic (e.g. Circle)


blue 21 62 80


blue 23 66 80 blue 27 70 80

orange 21 62 80

green 97 75 50 green 94 71 50

orange 21 62 80

PublisherSubscriber


OpenSplice DDS

From Shapes to Market Data


The Market Data TopicType Declaration

symbol name exchange quoteAAPLGOOGMSFT

Apple Inc. NASD 165.37Google Inc. NASD 663.97

Microsoft Corp. NASD 33.73

struct StockQuote { char<4> symbol;

char<4> exchange; char<64> name; float quote;

};#pragma keylist StockQuote symbol

Topic Instantiation

// Create Topic with Default QoSdds::Topic<StockQuote> topic(“stock_quote”);

// Create Content Filtered Topicdds::ContentFilteredTopic cfTopic = topic.create_content_filtered(expr, params);


Filtering Market Data

‣ DDS continuous queries can be used for filtering the stream of data produced for a specific stock symbol.

0

10

20

30

40

Microsoft

MSFT

154.0

156.5

159.0

161.5

164.0

Untitled 1 Untitled 20Untitled 39Untitled 58Untitled 77Untitled 96

Apple

AAPL

MSFT > $35

AAPL < $157 | AAPL > $162


Writing Market DataCreating a Writer


// Create a Writer with Default Qosdds::Writer<StockQuote> writer(topic);





Writing Data

StockQuote quote = { “AAPL”, “NASD”, 163.37, “Apple Inc.”};

writer.write(quote);





Writing Data


writer.write(quote);

Fastest Way of Writing Data


dds::InstanceWriter iwriter = writer.register_instance(quote);iwriter.write(quote);


OpenSplice DDS

Read vs. Takestruct Counter { int cID; int count;};#pragma keylist Counter cID

Topic Used in next section.

DataReader Cache

Topic

Pro

prie

tary

Info

rmat

ion

- D

istri

butio

n w

ithou

t Exp

ress

ed W

ritte

n P

erm

issi

on is

Pro

hibi

ted.


Reading Samples

‣ Read iterates over the available sample instances

‣ Samples are not removed from the local cache as result of a read

‣ Read samples can be read again, by accessing the cache with the proper options (more later)

DataReader

1 1

2 1

3 1

2 2

Samples Read Samples not Read

1 2 1 3 1 4

2 3

3 2 3 3 3 4 3 5

DataReader Cache

Topic

Pro

prie

tary

Info

rmat

ion

- D

istri

butio

n w

ithou

t Exp

ress

ed W

ritte

n P

erm

issi

on is

Pro

hibi

ted.


Reading Samples

‣ Read iterates over the available sample instances

‣ Samples are not removed from the local cache as result of a read

‣ Read samples can be read again, by accessing the cache with the proper options (more later)

Samples Read Samples not Read

DataReader

1 1

2 1

3 1

2 2

1 2 1 3 1 4

2 3

3 2 3 3 3 4 3 5

DataReader Cache

Topic

Pro

prie

tary

Info

rmat

ion

- D

istri

butio

n w

ithou

t Exp

ress

ed W

ritte

n P

erm

issi

on is

Pro

hibi

ted.


Taking Samples

‣ Take iterates over the available sample instances

‣ Taken Samples are removed from the local cache as result of a take

‣DataReader

1 1

2 1

3 1

2 2

Samples not Taken

1 2 1 3 1 4

2 3

3 2 3 3 3 4 3 5

DataReader Cache

Topic

Pro

prie

tary

Info

rmat

ion

- D

istri

butio

n w

ithou

t Exp

ress

ed W

ritte

n P

erm

issi

on is

Pro

hibi

ted.


Taking Samples



Samples not Taken

DataReader 2 2

1 2 1 3 1 4

2 3

3 2 3 3 3 4 3 5

DataReader Cache

Topic

Pro

prie

tary

Info

rmat

ion

- D

istri

butio

n w

ithou

t Exp

ress

ed W

ritte

n P

erm

issi

on is

Pro

hibi

ted.


Taking Samples



Samples not Taken

DataReader

1 3 1 4

2 3

3 3 3 4 3 5


Reading DataCreating a Reader


// Create a Reader with Default Qosdds::Reader<StockQuote> reader(topic);





Reading Data

StockQuoteSeq data;dds::SampleInfoSeq info;

reader.read(data, info);





Reading Data


reader.read(data, info);

Reading a Specific Instance

StockQuote quote;quote.symbol = “APPL”;StockQuoteSeq data;dds::SampleInfoSeq info;

dds::InstanceReader ireader = reader.instance(quote);ireader.read(data, info);


Taking DataCreating a Reader







Taking Data


reader.take(data, info);





Taking Data


reader.take(data, info);

Taking a Specific Instance

StockQuote quote;quote.symbol = “APPL”;StockQuoteSeq data;dds::SampleInfoSeq info;

dds::InstanceReader ireader = reader.instance(quote);ireader.take(data, info);

Pro

prie

tary

Info

rmat

ion

- D

istri

butio

n w

ithou

t Exp

ress

ed W

ritte

n P

erm

issi

on is

Pro

hibi

ted.


Sample, Instance and View States

‣ Along with data samples, DataReaders are provided with state information allowing to detect relevant transitions in the life-cycle of data as well as data writers

‣ Sample State (READ | NOT_READ): Determines wether a sample has already been read by this DataReader or not.

‣ Instance State (ALIVE, NOT_ALIVE, DISPOSED). Determines wether (1) writer exist for the specific instance, or (2) no matched writers are currently available, or (3) the instance has been disposed

‣ View State (NEW, NOT_NEW). Determines wether this is the first sample of a new (or re-born) instance

Dat

aRea

der

History Depth = 2

DataReader Cache

Samples

1 1

2 2

3 1

1 2

2 3

SampleInfo

1

2

3

1

2

Topic

Pro

prie

tary

Info

rmat

ion

- D

istri

butio

n w

ithou

t Exp

ress

ed W

ritte

n P

erm

issi

on is

Pro

hibi

ted.


Would you let me know...

DDS provides three main mechanism for exchanging information with the application

‣Polling. The application polls from time to time for new data or status changes. The interval might depend on the kind of applications as well as data

‣WaitSets. The application registers a WaitSet with DDS and waits (i.e. is suspended) until one of the specified events has happened.

‣ Listeners. The application registers a listener with a specific DDS entity to be notified when relevant events occur, such as state changes or

Pro

prie

tary

Info

rmat

ion

- D

istri

butio

n w

ithou

t Exp

ress

ed W

ritte

n P

erm

issi

on is

Pro

hibi

ted.


Reading with Waitsets// Functor handling new data arrivalstruct MarketDataHandler { void operator()(MarketDataReader& reader);};

1

Pro

prie

tary

Info

rmat

ion

- D

istri

butio

n w

ithou

t Exp

ress

ed W

ritte

n P

erm

issi

on is

Pro

hibi

ted.



1

// Create & Register ConditionMarketDataHandler functor;dds::ReadCondition cond = reader.create_read_condition(functor);dds::WaitSet ws;ws.attach(cond);

2

Pro

prie

tary

Info

rmat

ion

- D

istri

butio

n w

ithou

t Exp

ress

ed W

ritte

n P

erm

issi

on is

Pro

hibi

ted.



1


2

// Wait & Handledds::ConditionVector cvec = ws.wait();for (int i = 0; i < cvec.size(); ++i) cvec[i].execute();

3’

Pro

prie

tary

Info

rmat

ion

- D

istri

butio

n w

ithou

t Exp

ress

ed W

ritte

n P

erm

issi

on is

Pro

hibi

ted.



1


2

// Wait & Handledds::ConditionVector cvec = ws.wait();for (int i = 0; i < cvec.size(); ++i) cvec[i].execute();

3’ // Dispatchws.dispatch();

3”

Pro

prie

tary

Info

rmat

ion

- D

istri

butio

n w

ithou

t Exp

ress

ed W

ritte

n P

erm

issi

on is

Pro

hibi

ted.


Reading with Listeners

// Functor handling new data arrivalclass MarketDataListener {public: void handle_data(MarketDataReader& reader);};

1

Pro

prie

tary

Info

rmat

ion

- D

istri

butio

n w

ithou

t Exp

ress

ed W

ritte

n P

erm

issi

on is

Pro

hibi

ted.


Reading with Listeners

// Functor handling new data arrivalclass MarketDataListener {public: void handle_data(MarketDataReader& reader);};

1

// Functor handling new data arrivalMarketDataListener mdl;dds::connection cnt = reader.on_data_available_signal_connect( std::bind(&MarketDataListener::handle_data, &mdl, _1));

2


OpenSplice DDS

Quality of Service


QoS Model

‣ QoS-Policies are used to control relevant properties of OpenSplice DDS entities, such as:‣ Temporal Properties

‣ Priority

‣ Durability

‣ Availability

‣ ...

‣ Some QoS-Policies are matched based on a Request vs. Offered Model thus QoS-enforcement

‣ Publications and Subscriptions match only if the declared vs. requested QoS are compatible ‣ e.g., it is not possible to match a publisher which delivers data unreliably with a subscriber which requires reliability

Publisher

DataWriter

Topic

Type

QoS

Name

writes

QoS

DataWriter

Topic

Typewrites

Subscriber

DataReaderreads

DataReaderreads

...

QoS

Name

QoS

QoS QoS

QoS matching

......

QoS QoS

Type Matching

DomainParticipant DomainParticipant

QoS QoS


QoS Policies

‣ Rich set of QoS allow to configure several different aspects of data availability, delivery and timeliness

‣ QoS can be used to control and optimize network as well as computing resource

QoS Policy Applicability RxO Modifiable

DURABILITY

DURABILITY SERVICE

LIFESPAN

HISTORY

PRESENTATION

RELIABILITY

PARTITION

DESTINATION ORDER

OWNERSHIP

OWNERSHIP STRENGTH

DEADLINE

LATENCY BUDGET

TRANSPORT PRIORITY

TIME BASED FILTER

RESOURCE LIMITS

USER_DATA

TOPIC_DATA

GROUP_DATA

T, DR, DW Y N Data AvailabilityT, DW N N

Data Availability

T, DW - Y

Data Availability

T, DR, DW N N

Data Availability

P, S Y N Data DeliveryT, DR, DW Y N

Data Delivery

P, S N Y

Data Delivery

T, DR, DW Y N

Data Delivery

T, DR, DW Y N

Data Delivery

DW - Y

Data Delivery

T, DR, DW Y Y Data TimelinessT, DR, DW Y Y

Data Timeliness

T, DW - Y

Data Timeliness

DR - Y Resources

T, DR, DW N N

Resources

DP, DR, DW N Y ConfigurationT N Y

Configuration

P, S N Y

Configuration

Publisher

DataWriter

Topic

Type

QoS

Name

writes

QoS

DataWriter

Topic

Typewrites

Subscriber

DataReaderreads

DataReaderreads

...

QoS

Name

QoS

QoS QoS

QoS matching

......

QoS QoS

Type Matching

DomainParticipant DomainParticipant

QoS QoS

Pro

prie

tary

Info

rmat

ion

- D

istri

butio

n w

ithou

t Exp

ress

ed W

ritte

n P

erm

issi

on is

Pro

hibi

ted.


Mapping QoS

Data Latency

LatencyBudget TransportPriority

Throughput

DeadlineTimeBasedFilter

Control over Latency/Throughput tradeoff Control over data latency Control over data priority

Which properties does QoS controls?

Pro

prie

tary

Info

rmat

ion

- D

istri

butio

n w

ithou

t Exp

ress

ed W

ritte

n P

erm

issi

on is

Pro

hibi

ted.


Mapping QoS

Data Latency


Throughput



Data Availability

OwnershipOwnershipStrength

Durability

History

Lifespan

Control over data queueing Control over data persistency Control over data sources hot-swap


Pro

prie

tary

Info

rmat

ion

- D

istri

butio

n w

ithou

t Exp

ress

ed W

ritte

n P

erm

issi

on is

Pro

hibi

ted.


Mapping QoS

Data Latency


Throughput



Data Availability

OwnershipOwnershipStrength

Durability

History

Lifespan

Control over data queueing Control over data persistency Control over data sources hot-swap

Data Delivery

Reliability

DestinationOrder

Presentation

Control over data distribution reliability Control over data ordering Control over presentation

OpenSplice DDS provides programmatic QoS-driven support for configuring the most important properties of data distribution!



OpenSplice DDS

Playing with QoS

Pro

prie

tary

Info

rmat

ion

- D

istri

butio

n w

ithou

t Exp

ress

ed W

ritte

n P

erm

issi

on is

Pro

hibi

ted.


History QoS

‣ The History QoS Controls the number of samples-per-instance that will be stored by the middleware on behalf of a Reader

‣ Keep Last K. The History QoS can be set so to always have the latest K samples

‣ Keep All. The History QoS can be set so keep all samples produced by the writer and not yet taken, until resource limits are not reached

DataReader

History Depth = 5

Topic

1 1

2 1

3 1

1 2 1 3 1 4 1 5

2 2 2 3 2 4 2 5

3 2 3 3 3 4 3 5

// Create Reader QoSdds::DataReaderQos drqos;drqos.set_keep_last(N);// Create Readerdds::DataReader<MarketData> reader(topic, drqos);

// Create Reader QoSdds::DataReaderQos drqos;drqos.set_keep_all();// Create Readerdds::DataReader<MarketData> reader(topic, drqos);

1

2

Pro

prie

tary

Info

rmat

ion

- D

istri

butio

n w

ithou

t Exp

ress

ed W

ritte

n P

erm

issi

on is

Pro

hibi

ted.


Transport Priority QoS

// Create Writer QoSdds::DataWriterQos dwqos;dwqos.set_priority(MID_PRIO);// Create Writerdds::DataWriter<MarketData> writer(topic, dwqos);// Write Datawriter.write(data);// Change Prioritydwqos.set_priority(HIGH_PRIO);writer.writer(data)

1

Make sure that important data gets the right priority

Traffic Shaping

OpenSplice DDSBinding

Networking

Shared

Memory

OpenSplice DDSBinding

Networking

Shared

Memory

Network Channels

Priority Bands

Pre-emptive Network Scheduler

Priority Scheduler

Data Urgency Traffic Pacing

Single Copy per Node

Pack Across Topics/Applications

Optimal Unmarshaling

Pro

prie

tary

Info

rmat

ion

- D

istri

butio

n w

ithou

t Exp

ress

ed W

ritte

n P

erm

issi

on is

Pro

hibi

ted.


Ownership QoSAvailability of data producers can be controlled via two QoS Policies

‣ OWNERSHIP

‣ OWNERSHIP STRENGTH

‣ Instances of exclusively owned Topics can be modified (are owned) by the higher strength writer

‣ Writer strength is used to coordinate replicated writers

R1

R2

R3

symbol: "MSFT"name: "Microsoft Corp."exchange: "NASD"quote: 33.73

StockQuote

symbol: "GOOG"name: "Google Inc."exchange: "NASD"quote: 663.97

StockQuote

W1

W2’’

STRENGTH=3

STRENGTH=1


StockQuote


StockQuotesymbol: "AAPL"name: "Apple Inc."exchange: "NASD"quote: 165.37

StockQuote

W1’

STRENGTH=2


StockQuote



StockQuote


StockQuote


StockQuote

symbol: "AAPL"name: "Apple Inc."exchange: "NASD"quote: 165.37


StockQuote


StockQuote

Pro

prie

tary

Info

rmat

ion

- D

istri

butio

n w

ithou

t Exp

ress

ed W

ritte

n P

erm

issi

on is

Pro

hibi

ted.



‣ OWNERSHIP




R1

R2

R3


StockQuote


StockQuote

W1

W2’’

STRENGTH=3

STRENGTH=1


StockQuote


StockQuote

W1’

STRENGTH=2


StockQuote


StockQuote


StockQuote


StockQuote


StockQuote


StockQuote


StockQuote

Pro

prie

tary

Info

rmat

ion

- D

istri

butio

n w

ithou

t Exp

ress

ed W

ritte

n P

erm

issi

on is

Pro

hibi

ted.



‣ OWNERSHIP




R1

R2

R3


StockQuote


StockQuote

W1

W2’’

STRENGTH=3

STRENGTH=1


StockQuote

W1’

STRENGTH=2


StockQuote


StockQuote


StockQuote


StockQuote


StockQuote


StockQuote

Pro

prie

tary

Info

rmat

ion

- D

istri

butio

n w

ithou

t Exp

ress

ed W

ritte

n P

erm

issi

on is

Pro

hibi

ted.


Ownership QoS Availability of data producers can be controlled via two QoS Policies

‣ OWNERSHIP





StockQuote


StockQuote


StockQuote


StockQuote


StockQuote


StockQuote


StockQuote


StockQuote


StockQuote

R1

R2

R3W2’’

STRENGTH=1

Pro

prie

tary

Info

rmat

ion

- D

istri

butio

n w

ithou

t Exp

ress

ed W

ritte

n P

erm

issi

on is

Pro

hibi

ted.


Ownership QoS Availability of data producers can be controlled via two QoS Policies

‣ OWNERSHIP





StockQuote


StockQuote


StockQuote


StockQuote


StockQuote


StockQuote


StockQuote


StockQuote


StockQuoteR1

R2

R3W2’’

STRENGTH=1


OpenSplice DDS

Online References

http://www.opensplice.org/

http://www.opensplice.com/


Pro

prie

tary

Info

rmat

ion

- D

istri

butio

n w

ithou

t Exp

ress

ed W

ritte

n P

erm

issi

on is

Pro

hibi

ted.


SIMD Project

http://code.google.com/p/simd-cxx/




OpenSplice DDS

Summing Up

Pro

prie

tary

Info

rmat

ion

- D

istri

butio

n w

ithou

t Exp

ress

ed W

ritte

n P

erm

issi

on is

Pro

hibi

ted.


Concluding Remarks‣ DDS is the ideal fit for addressing several of the

data distribution and management challenges faced by financial firms, such as:‣ Market Data Distribution

‣ High Performance Caching

‣ Real-Time Business Intelligence

‣ OpenSplice DDS is an Open Source implementation of the DDS standard widely deployed and delivering great performance

‣ OpenSplice DDS is Commercially Supported Open Source Software

Delivering Performance, Openness, and Freedom

OpenSplice DDS

Enterprise Ed.

Community Ed.

Compact Ed.

Professional Ed.

Embrace Performance, Openness and Freedom. Join the OpenSplice DDS Community at www.opensplice.org

http://www.opensplice.org

http://www.opensplice.org

Online Resources


emailto:[email protected]

http://www.youtube.com/OpenSpliceTube http://opensplice.blogspot.com

http://bit.ly/1Sreg


http://www.slideshare.net/angelo.corsaro

http://twitter.com/acorsaro/

http://www.prismtech.com/opensplice-dds








http://bit.ly/1Sreg

http://bit.ly/1Sreg





i need the data and i need it 1ms ago!

Technology

datadistribution challenges

data aggregators

data feeds challenges

dds sig

trading latency market

key dds vendors

market data angelo corsaro

subscription api