Christian Esposito 16 th July 2008 Flexible Communication Among DDS Flexible Communication Among DDS Flexible Communication Among DDS Flexible Communication Among DDS Publishers and Subscribers Publishers and Subscribers Christian Esposito and Domenico Cotroneo Università di Napoli Università di Napoli Federico II Federico II Dario Di Crescenzo Federico II Federico II christian esposito@unina it www mobilab unina it Consorzio SESM Consorzio SESM Soluzioni Innovative Soluzioni Innovative christian.esposito@unina.it www.mobilab.unina.it
29
Embed
Flexible Communication Among DDSFlexible ......JSON and YAML parsers are more efficient than DOM XML, and SAX XML has performance comparable to YAML when www mobilab unina it christian
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Christian Esposito
16th July 2008
Flexible Communication Among DDSFlexible Communication Among DDSFlexible Communication Among DDS Flexible Communication Among DDS Publishers and SubscribersPublishers and Subscribers
Christian Esposito and Domenico CotroneoUniversità di NapoliUniversità di Napoli
The present work is the result of the collaborationb M bil b R h G (U i i à di N li)between Mobilab Research Group (Università di Napoli)and Selex-SI under the project “IniziativaSoftware”(http://www.iniziativasoftware.it).
::.. Problem Statement(1/2)Ult L S l (ULS) t h t i d bUltra-Large-Scale (ULS) systems are characterized by afederation of heterogeneous and autonomous systemsinterconnected by wide-area and unmanaged networks.
Due to geographical extension andpolitical reasons there is no centralpolitical reasons, there is no centralcontrol in ULS systems
It is possible to have a non-uniformevolution of the system: modifiedand not-modified applications have
christian esposito@unina itwww mobilab unina it
and not modified applications haveto coexist (multi-versioning).
The PSM over UDP of RTPS specificationindicates Common Data Representation (CDR)as Wire Representation for the exchanged
CDR couples the comprehension of the received message to
as Wire Representation for the exchangedmessages.
CDR couples the comprehension of the received message tothe knowledge of the sent message structure. Therefore, ifthe publisher changes the message structure, theunchanged subscribers cannot understand the receivedunchanged subscribers cannot understand the receivedmessage.
1. Definition of other serialization formats that allowguaranteeing multi-versioning property;g g g p p y;
2. Realization of a DDS-based prototype that adoptsthese formats instead of CDR to serialize andthese formats instead of CDR to serialize anddeserialize application data;
3 Assessment of the performance achievable with these3. Assessment of the performance achievable with theseformats;
To obtain flexibility, CDR has to be replaced with a tree-based serialization formats:
The publisher disseminates on the wire not only theThe publisher disseminates on the wire not only themessage content but also a representation of themessage structure as a tree so to overcome the CDRlimitslimits.
XML is widely adopted to facilitate the sharing oft t d d t diff t tstructured data across different systems,
particularly in a wide area scenario.
christian esposito@unina itwww mobilab unina it
Message structure is coded into the opening andclosing tags that wrap the data.
::.. Solution (2/2)XML syntax is redundant or large relative to binaryrepresentations (e.g. CDR) of similar data, and this affectsapplication efficiency through higher transmission andpp y g gserialization costs.
M ffi i t i li ti f t d dMore efficient serialization formats are needed
JSON and YAML are two tree-basedformats that offer the same messagestructure coding as XML but adopt a
christian esposito@unina itwww mobilab unina it
JSONstructure coding as XML, but adopt amechanism of name/value pair thatallows saving bytes.
A measurement-based campaign has been conducted usingthe DDS solution provided by RTI and these parsers:
• CDR: In-house developed parser;• XML: XERCES (both in DOM and SAX mode);XML: XERCES (both in DOM and SAX mode);• JSON: JSON-C (http://oss.metaparadigm.com/json-c/)• YAML: LIBYAML (http://pyyaml.org/wiki/LibYAML)
Scope: evaluate the performance when using tree-basedi li i f d i h h CDR
christian esposito@unina itwww mobilab unina it
serialization formats and compare with the CDR case.
The solution to improve the efficiency of the tree-basedformats is using a data compression technique, so toreduce the bytes exchanged over the network.
Scope: measure the performance of the DDS product whenusing data compression and assess the achievableg pimprovement in terms of stream size and latency.
Metrics:Metrics:• Compression ratio
χ = serialization_stream_size / compr_streamC i S i li i ffi i• Compression+Serialization efficiency
::.. Performance Evaluation III (2/4)The realistic behaviour of a typical wide-The realistic behaviour of a typical wide-area network is characterized throughtraffic measuring over several paths onPlanetLabPlanetLab.
ScenarioLow
S i
Three scenarios are defined
ScenarioHigh
christian esposito@unina itwww mobilab unina it
Three scenarios are definedfrom the obtained measures.