Efficient XML Interchange

1
Efficient XML Interchange
2
XML

• Why is XML good?
• A widely accepted standard for data
  representation
• Fairly simple format
• Flexible
• Its not used by everyone, but its used by
  enough people to make for a rich tools
  environment
• Its flexible enough to be used in lots of
  contexts
• Its text based and human readable, which makes
  it a good archival format

3
XML

• XML in 10 points
• http//www.w3.org/XML/1999/XML-in-10-Points
• Includes (3) XML is meant to be read, and (4)
  XML is verbose by design
• XML can (but should not be) read by humans, and
  is not very compact

4
XML

• These design principles also make it very
  difficult to use XML in some environments
• Wireless military links low bandwidth
• Mobile devices battery life limitations
•  Processing efficiency it can take CPU cycles
  to parse XML
• Data binding

5
Limitations

• Lots of ships have 64 Kbit/sec at best. It is
  problematic to ship XML across these links
• CPUs are on Moores law curve, but battery power
  is limited by the state of chemistry. We cant
  assume that faster processors will save us. Lots
  of applications for hand held devices with
  limited battery power (cell phones, etc.)
• Cell phones dont necessarily have strong CPUs,
  so parsing XML can be expensive relative to other
  tasks

6
Data Binding

• This is a more subtle problem.
• ltPoint x1.0 y2.0/gt
• How do you convert this to an object? You need to
  parse the string 1.0, then convert it to a
  binary representation
• Its the difference between
• string x
• And
• float x

7
Data Binding

• Typically something comes in from the wire, and
  you have to do the Java equivalent of
• Float.parseFloat(1.0)
• This is expensive when working with numeric-heavy
  documents
• It is much more efficient to keep the value X in
  a binary representation in the document, then
  simply read it on the receiving side

8
Efficient XML Interchange

• EXI relaxes some of the requirements of XML in
  order to be more compact, faster to parse, and
  have better data binding characteristics
• Relax the human readable requirement
• Allow binary data
• What you get is an alternate encoding of the XML
  infoset that is more compact, faster to parse,
  and allows deployment in new environments that
  XML previously could not be deployed in

9
EXI

• EXI is being developed by a W3C working group and
  is on a standards track. The hope is that this
  will become a W3C-blessed encoding of the XML
  infoset
• Working group draft now working its way to
  approval.
• Need multiple implementations, blessed by W3C
  technical architecture group, approval by other
  W3C working groups (encryption, processors, etc.)

10
EXI

• Represents the same data as an XML document,
  only in a more efficient encoding
• Minimal impact on other XML technologies, such
  as encryption
•  More efficient to parse, better data binding
  performance

11
EXI

• http//www.w3.org/XML/EXI
• Includes file format specification, primer on
  EXI, best practices
• Note that one thing that is NOT specified is an
  API for accessing the data. This is an important
  and significant omission
• Lack of a standardized typed API means we still
  have to go through string representations

12
Typed API

• What is meant by a typed API?
• DOM and SAX return string values
• Attr anAttribute
• // DOM returns a String attribute value here
• String val anAttribute.getValue()
• And then we need to convert val into a float via
• Float aFloat Float.parseFloat(val)

13
Typed API

• But what we often want is the value specified in
  the schema
• Float aFloat anAttribute.getFloat()
• There are proposals for a generalized typed API,
  but it is not part of this standard

14
EXI

• EXI has several options to handle different
  situations.
• You have an XML document and a schema
• You have an XML document but no schema
• You have an XML document, and a schema that
  almost, but not quite, matches the document

15
Element and Attribute Names

• Tag names take up a lot of space, and can be
  somewhat expensive to parse
• ltName firstJames lastMadisongt
• ltStategtVirginialt/Stategt
• lt/Namegt
• Count up the characters used for markup
  here31/55 50-60 of file size for markup tags
• If we replace the character tags with numeric
  stand-ins we can get much more compact, and it
  will be faster to parse

16
Schema-Informed

• If you have a schema, that gives you type
  information about the XML document. You know that
  ltfoo x1.0/gt means the x is a float value
  rather than a string, because the schema tells
  you that.
• That means you can store the 1.0 value in a
  binary format, which is generally more compact
  and has the potential to have better data binding
  with a typed API

17
Schemaless

• What if you dont have a schema? This means you
  cant exploit type information. But EXI should
  support this situation, because it should be a
  general solution
• EXI handles this by replacing repeating strings
  with a compact identifier

18
Schemaless

• ltAddress townMonterey zip93943/gt
• The strings Monterey and the zip code are
  likely to be repeated many times in an XML
  document. We can create a table of these values,
  and then use the table ID rather than the whole
  string

String ID
Monterey 1
93940 2
San Jose 3
98842 5
19
Almost Schemas

• If you have a document that doesnt quite match
  the schema, EXI can take a forgiving attitude. It
  uses the schema to encode the types it knows
  about, and uses strings and string table
  identifiers to handle the ones not described by
  the schema

20
Implementations

• As of now there is one implementation of the
  draft spec, Efficient XML from Agile Delta
  (http//www.agiledelta.com)
• Other open source projects underway, and some
  commercial projects
• The standards process requires that multiple
  independent implementations be available before
  the standard is approved

21
Results

• Example Distributed Interactive Simulation (DIS)
  is an IEEE standard for modeling and simulation.
  It is a binary standard that contains (x,y,z),
  velocity, acceleration, and other numeric-heavy
  data
• We did an XML representation of the binary DIS
  standard

22
Results
DIS Binary (bytes) DIS XML EXI Format
1 PDU 144 1167 129
1000 PDUs 464,480 3,924,680 365,564
23
Results

• Somewhat better size than the original binary
  format. The exact size varies somewhat depending
  on the numeric data, while the original binary
  format is always the same size. Exi seems to be
  consistently better, though
• AND it is marked up in a way that makes it
  equivalent to an XML file. This means we can
  easily access all the tools of the XML ecosystem
  by simply converting it to a text XML
  representation

24
Conclusions

• Replace all text XML with EXI? No! EXI is
  intended to expand the use of XML into use cases
  that XML could not service. XML mostly does fine
  in its existing environment
• EXI can be used to XML-ify existing binary
  protocols and get slightly better performance
  with greatly increased interoperability (no one
  knows DIS binary, everyone knows XML)
• Next great frontier typed XML APIs