Filter-enabled Binary XML Dissemination in Embedded Networks
Sebastian K
¨
abisch
1,2
and Richard Kuntschke
1
1
Siemens AG, Corporate Technology, 81730 Munich, Germany
2
University of Passau, Chair of Distributed Information Systems, 94032 Passau, Germany
Keywords:
Binary XML, EXI, Embedded Networks, XPath, Filtering
Abstract:
Efficient data dissemination in distributed systems is a challenge that can be tackled by sharing common data
and processing results among multiple queries. Doing so in an effective manner helps to save network band-
width and computational resources. This is especially important in embedded networks where such resources
are often extremely scarce. Disseminating resource-intensive XML data in embedded networks has been en-
abled by using binary XML technologies such as W3C’s EXI format. In this paper, we show how filter-enabled
binary XML dissemination in embedded networks helps to further reduce resource demands. Thus, through
the suitable placement of pre- and post-filters on binary XML data, bandwidth on network connections and
computational resources on nodes can be saved. Consequently, more data can be processed with a certain
amount of available resources within an embedded network.
1 INTRODUCTION
Efficient data dissemination is an important task in
distributed systems where data needs to be trans-
ferred from data sources to data sinks located at dif-
ferent places within the system. Sharing common
data and processing results among multiple data sinks
helps to save network bandwidth and computational
resources. While such approaches reduce resource us-
age in any distributed system, their usage is crucial in
embedded networks (fundamental in, e.g., building or
industrial automation, automotive industry, and smart
grid) due to the strictly limited resources such as
from microcontrollers. Using binary XML techniques
such as W3C’s Efficient XML Interchange (EXI) for-
mat (Schneider and Kamiya, 2011) enables the dis-
semination of otherwise resource-intensive XML data
in embedded networks. Additionally, using filter-
enabled binary XML dissemination helps to further
reduce resource demands.
A filter-enabled subscription mechanism in em-
bedded networks reduces network traffic and unnec-
essary message processing at the client nodes. It opti-
mizes data interaction between the service provider
(data source) and the service requester/client (data
sink) in terms of data novelty and supports an effi-
cient execution of applications in embedded networks
at runtime. Consequently, more data can be processed
with a certain amount of available resources within an
Class:
Proce:
0
3
1
Class:
Proce:
0
1
2
Class:
Proce:
1
3
3
Class:
Proce:
1
2
5
Class:
Proce:
0
3
6
Class:
Proce:
0
3
7
Class:
Proce:
1
2
4
Class:
Proce:
1
3
8
1
1
1 1
1
1
1
1
1
Figure 1: Example embedded network with a service
provider (node 1) and three clients (node 3, 4, and 7)
embedded network.
An immediate evaluation at the node of service
data origin prevents the dissemination of data that
is outside of the scope of the corresponding service
requesters. In the context of constrained embedded
networks, however, a desired early evaluation at the
node of service data origin sometimes cannot be re-
alized. Two aspects support this observation: First of
all, the available resources, especially when it comes
to memory, are often not sufficient for installing an
additional filter application. Secondly, vendors of em-
40
Käbisch S. and Kuntschke R..
Filter-enabled Binary XML Dissemination in Embedded Networks.
DOI: 10.5220/0004852100400049
In Proceedings of the 10th International Conference on Web Information Systems and Technologies (WEBIST-2014), pages 40-49
ISBN: 978-989-758-023-9
Copyright
c
2014 SCITEPRESS (Science and Technology Publications, Lda.)
Class:
Proce:
0
3
1
Class:
Proce:
0
1
2
Class:
Proce:
1
3
3
Class:
Proce:
1
2
5
Class:
Proce:
0
3
6
Class:
Proce:
0
3
7
Class:
Proce:
1
2
4
Class:
Proce:
1
3
8
1 1
1
//value=27.0
//value<21.0
//value>20.5
...
<value>...
...
EXI
(a)
Class:
Proce:
0
3
1
Class:
Proce:
0
1
2
Class:
Proce:
1
3
3
Class:
Proce:
1
2
5
Class:
Proce:
0
3
6
Class:
Proce:
0
3
7
Class:
Proce:
1
2
4
Class:
Proce:
0
3
8
1 1
1
1
//value=27.0
//value<21.0
//value>20.5
...
<value>...
...
EXI
8
(b)
Figure 2: (a) Naive approach for data dissemination (b) Desired pre-/post-filter approach for data dissemination purposes.
bedded nodes do not offer the possibility of installing
such filter applications.
In this paper, we introduce an approach for orga-
nizing filter-based service data dissemination in con-
strained embedded networks that takes into account
resources such as device classes, device processing
performance, and connection quality between nodes.
The goal is to share relevant service data using bi-
nary XML whenever possible by using filters and sub-
filters, respectively. This reduces both network traffic
and computational load on nodes. Basically, this ap-
proach leads to content-based routing at the applica-
tion level based on binary XML content.
1.1 Problem Statement
Consider Figure 1 as an example embedded net-
work that shows a simple distributed system with
eight nodes and corresponding network connections.
Nodes are categorized here into three classes (1 to
3) indicating their capabilities. The lower the class
number, the more powerful the device. Class 1,
e.g., corresponds to a consumer PC, class 2 corre-
sponds to a hardware system with a resource com-
plexity found in home network routers, and class 3
corresponds to constrained embedded hardware such
as a microcontroller (e.g., ARM Cortex-M3
1
with
24MHz, 256kB RAM, and 16kB ROM). The boolean
proce value indicates whether there are computational
resources available at the corresponding node. A
value of 0 indicates that there are no resources left
for installing a potential application, while a value
1
http://www.arm.com/products/processors/cortex-
m/cortex-m3.php
of 1 indicates available resources. Resources in this
context may comprise memory as well as process-
ing power. The numbers at the network connec-
tions indicate the available connection quality. Val-
ues closer to 1 indicate superior connection quality.
In contrast, values closer to 0 indicate poor connec-
tion quality. Connection quality may be impacted by
the technology of the underlying physical communi-
cation link, e.g., low-power wireless communication
(e.g., IEEE 802.15.4 (IEEE, 2011)) vs. Ethernet, as
well as by the current system state as indicated, e.g.,
by the current package loss ratio, delay, and available
bandwidth of the communication link. Increasing net-
work traffic on a communication link lowers the con-
nection quality since, e.g., increasing traffic reduces
the available bandwidth.
Now, let us assume a simple data dissemination
scenario. In Figure 2(a), node 1 is the data source,
sending XML-based data including a value informa-
tion in EXI format, and nodes 3, 4, and 7 are the
data sinks that are interested in the value information.
The data sink at node 3 is only interested in data ele-
ments containing the value 27.0, node 4 is interested
in data elements containing values less than 21.0, and
node 7 is interested in data elements containing values
greater than 20.5. Since node 1 does not have the op-
portunity to set up a filter mechanism (proce is equal
to 0) to test client’s relevance of a new data message,
this node always sends an individual copy of the en-
tire data message to each data sink and the nodes con-
stituting the data sinks process the data accordingly
(see Figure 2(a)). However, this leads to relatively
high bandwidth usage on the affected network con-
nections and requires considerable computational re-
Filter-enabledBinaryXMLDisseminationinEmbeddedNetworks
41
sources at the data sink nodes, even if this message is
not relevant anyway. To illustrate the impact of this
naive dissemination approach in terms of connection
quality, in Figure 2(a) in each involved dissemination
link the connection quality value is decremented by
the value
1
10
.
Now consider Figure 2(b), showing our pre-/post-
filter approach for the same distributed network. All
produced data is sent out by the data source only once.
A pre-filter (G
F
) at node 5 filters the data of relevance
for nodes 3 and 8. A message for node 3 is only for-
warded when the value in the data message is equal
to 27.0. Node 8 receives only a message from node
5 when a value information is present in the message.
A post-filter (G
0
F
) at node 8 is applied to evaluate the
relevance for nodes 4 and 7. Consequently, node 8
forwards only the messages to node 4 when the value
is smaller than 21.0. Node 7 only receives a message
when the value is greater than 20.5.
Thus, compared to the naive approach, the re-
quested data is transmitted to each data sink using
less network bandwidth and less processing power
overall in the system, since processing results can be
shared among multiple data sinks. Also, we are able
to distribute the computational load across capable
nodes within the distributed system. We accept this
approach even if the resources of a small number of
node devices may be claimed by the filter placement.
This is seen by the processablity of node 8 that is set
to 0.
1.2 Paper’s Contributions and Outline
This paper presents the following contributions:
• We shortly introduce a previously developed fil-
ter mechanism based on W3C’s EXI format (Sec-
tion 2). This mechanism constitutes the basis for
our further work on filter-enabled binary XML
dissemination in embedded networks.
• In Section 3, we present our new approach for ef-
ficient filter-enabled binary XML data dissemina-
tion in embedded networks. The approach helps
to reduce bandwidth usage on network connec-
tions and computational load on nodes, thus al-
lowing larger amounts of data to be transferred
and processed with a certain amount of available
resources.
• Finally, we present evaluation results showing the
effectiveness of our approach (Section 4).
2 EXI FILTER MECHANISM
In this section we give a short introduction about the
EXI format and the basic idea, how an EXI grammar
can be transformed to a filter grammar for evaluating
XPath expressions.
2.1 The W3C EXI Format
W3C, the inventor and standardizer of XML, faced
the drawbacks of plain-text XML and created a
working group called XML Binary Characterization
(XBC) (Goldman and Lenkov, 2005) to analyze the
condition of a binary XML format that should also
harmonize with the standardized plain-text XML for-
mat as well as with the XML Infoset. The outcome
was the start of the W3C Efficient XML Interchange
(EXI) format, which gained recommendation status at
the beginning of 2011 (Schneider and Kamiya, 2011).
Mainly, EXI is a grammar driven approach that is ap-
plied to bring XML-based data into a binary form and
vice versa. Such a grammar is constructed based on
a given XML Schema where each defined complex
type is represented as a deterministic finite automa-
ton (DFA). Figure 3(a) shows an excerpt of a sample
EXI grammar (set of automaton) G that can be used
for encoding and decoding. This grammar reflects an
XML Schema including the SOAP framework (Gud-
gin et al., 2003) with a status information within the
Header part and request/response patterns of temper-
ature and humidity information within the Body part.
Please note that the Root grammar is a predefined
grammar that occurs in each EXI grammar represen-
tation of arbitrary XML Schemas. It contains all entry
points of all root elements in a given schema. Here,
we only highlight in our context the relevant root En-
velope of the SOAP message framework. In general,
each DFA contains one start state and one end state,
which reflect the beginning and the end, respectively,
of a complex type declaration. Transitions to the next
state represent the sequential order of element and/or
attribute declarations within a complex type. Optional
definitions (e.g., choice, minOccurs = 0, etc.) are re-
flected by multiple transitions and assigned an event
code (EV). E.g., the SOAP Envelope message frame-
work embeds an optional Header and a mandatory
Body element (Gudgin et al., 2003). The equiva-
lent EXI grammar representation, as can be seen in
Figure 3(a) (Envelope Grammar), provides two tran-
sitions from the start state: one to the Header state
and one to the Body state. For signalization, a one
bit event code is used and assigned to the transition
(EV(1) for the Header; EV(0) for the Body). Gener-
ally, the number of bits used for m transitions is deter-
WEBIST2014-InternationalConferenceonWebInformationSystemsandTechnologies
42
Envel-
ope
Root Grammar
EV(000)
Header
EV(0)
EV(1)
EV(-)
Body
EV(-)
Envelope Grammar
EV(-)
status
EV(-)
Header Grammar
get-
Hum.
Body Grammar
Temperature Grammar
scale
EV(0)
EV(1)
EV(-)
value
EV(-)
EV(00)
Temp-
erature
EV(-)
get-
Temp.
Hum-
idity
EV(01)
EV(10)
EV(11)
EV(-)
EV(-)
EV(-)
Humidity Grammar
(a)
Envel-
ope
Root Grammar
EV(000)
Header
EV(0)
EV(1)
Body
EV(-)
Envelope Grammar
EV(-)
status
EV(-)
Header Grammar
Body Grammar
Temperature Grammar
scale
EV(0)
value
EV(-)
AS={Q }
2
PS={Q , Q }
3
<21.0
PS={Q }
2
='OK'
AS={Q , Q }
3
EV(11)
Temp-
erature
>20.5
4
4
EV(10)
Hum-
idity
AS={Q }
1
EV(1)
(b)
Figure 3: (a) EXI grammar G (excerpt) for encoding and decoding purposes (b) Filter Grammar G
F
with G
F
⊆ G based on
the queries Q
1
, Q
2
, Q
3
, and Q
4
for evaluation purposes.
mined by dlog
2
me. EV(-) on transitions indicates, no
event code is required.
An example XML message snipped such as
<Envelope><Header><status>OK< /status>...
would be transformed to a
000 1 ’OK’ ...
EXI (bit-)stream based on the EXI grammar shown
in Figure 3(a). This already shows, how compact
EXI can be. In addition, EXI is a type aware en-
coder that provides efficient coding mechanisms for
the most common data types (int, float, enumerations,
etc.). There are use cases in which the EXI repre-
sentation is said to be over 100 times smaller than
XML (Bournez, 2009). Based on the high compres-
sion ratio and the opportunity to obtain the data con-
tent directly from the EXI stream, XML-based mes-
saging is also feasible in the embedded domain, even
if constrained devices are used (K
¨
abisch et al., 2011).
2.2 EXI Filtering
EXI grammars build the bases for writing and read-
ing binary XML data. Previous work (K
¨
abisch et al.,
2012) shows the functionality to create an efficient
filter mechanism for binary XML data based on a
number of service requesters by providing XPath ex-
pressions that address the desired service data occur-
rences and/or data value conditions. Two approaches
which are feasible to constrained devices such as mi-
crocontrollers were presented: BasicEXIFiltering and
OptimizedEXIFiltering. The BasicEXIFiltering oper-
ates on top of an EXI grammar and evaluates normal-
ized XPath queries by means of binary XML. Opti-
mizedEXIFiltering presents a more sophisticated ap-
proach; it maps all XPath expressions within an EXI
grammar and removes all states and transitions which
are not required for message evaluation. The outcome
is a filter grammar denoted as G
F
. Figure 3(b) shows
such a constructed filter grammar G
F
based on the
queries
Q
1
= //Humidity
Q
2
= //status[text() =
0
OK
0
]
Q
3
= //Temperature/value[text() < 21.0]
Q
4
= //Temperature/value[text() > 20.5]
applied on the data model represented by the EXI
grammar G in Figure 3(a).
Applying the previously created EXI stream
000 1 ’OK’ ...
to the filter grammar G
F
we would, at least, success-
fully find a match for the XPath expression Q
2
. This
is due to the remaining states and transitions in G
F
that lead via the Envelope state (reading the bits 000
from the stream) and the Header state (reading the bit
1 from the stream) to the Header status state which
is dedicated as a predicate state (PS) and as an ac-
cepting state (AS). A predicate state results to a pred-
icate evaluation. Here, we have to evaluate whether
the message contains the status value ’OK’ which is
also the case in our example. An accepting state rep-
Filter-enabledBinaryXMLDisseminationinEmbeddedNetworks
43
resents a query match of a particular query (here Q
2
).
This is only true, if all aforegoing relevant predicate
states resulted in a positive predicate evaluation.
The next section explains how such filter gram-
mars can be used to realize an efficient binary XML
data dissemination mechanism in constrained embed-
ded networks.
3 DISSEMINATION ALGORITHM
This section presents our filter-enabled binary XML
dissemination algorithm. First of all, we introduce
our cost model and formalize the optimization prob-
lem (Section 3.1). Next, the dedicated algorithm is
presented (Section 3.2). Finally, an example is con-
sidered for better clarification of how our algorithm
works (Section 3.3).
3.1 Cost Model
Before we formalize our cost model we will formally
define an embedded network in our context as de-
scribed in Section 1.1 by N
emb
= (V,E,c,w, p). V de-
scribes the set of vertexes/nodes. Set E describes the
set of edges/connections between two nodes. Func-
tion c with c : V → N
>0
associates the device class of
a device node. The weight function w with w : E →
R
[0,1]
describes the connection quality between two
device nodes and p with p : V → {0,1} the process-
ing capability of a node.
A newly installed application, including a service
provider node (v
s
) and service subscriber clients
(C = {v
c
1
,...,v
c
n
}), in an embedded network N
emb
would typically lead to additional network traffic
and processing costs. To keep this overhead as
small as possible, we filter the data of relevance and
share this data as long as possible on a determined
dissemination path that avoids constrained device
class nodes and uses connections with relatively
good quality. Consequently, we have two metrics
which have to be considered: device class with the
processability (c and p) and connection quality (w).
Putting this together, we define our cost function f
for a given N
T
emb
= (V
T
,E
T
,c,w, p) with N
T
emb
⊆ N
emb
that only contains the nodes and transitions (spans a
tree) from v
s
to all nodes in C that is used for the data
dissemination:
f (N
T
emb
) := α ·
∑
v
i
∈V
T
(c(v
i
) + 1 − p(v
i
))
+(1 − α) ·
∑
(v
i
,v
j
)∈E
T
1
w(v
i
,v
j
)
.
(1)
The first summand formalizes the device class
with the processing capability of a device node. A
hop-noise value 1 is added to enable an influential de-
cision when we also have only one class occurrences
(c = 1) and each has processing capability (p = 1) in
the network. The second summand represents the re-
ciprocal connection quality. The α ∈ [0, 1] is a weight
factor that enables us to set up a more dominant part
in the cost function: the device class (α > 0.5 ) or the
connection quality (α < 0.5).
Using f we are able to formalize our optimization
problem to find a subgraph N
T
emb
of N
emb
for a filter-
enabled service data dissemination:
Minimize f
T
(2)
subject to
∑
v
i
∈V
T
p
T
(v
i
) ≥ 1 .
The inequality constraint specifies the occurrence
of at least one node processing capability within N
T
emb
that can be used to set up a pre-filter.
Unfortunately, for any constellation in N
emb
we
are not able to find an optimized solution in polyno-
mial time. In the next section we are going to present
an heuristic approach based on greedy algorithms that
approximates an optimized N
T
emb
for a filter-enabled
service data dissemination.
3.2 Algorithm
We are now going to describe our filter-enabled
service data dissemination algorithm, the
FilterEnabledDissemination algorithm (see Al-
gorithm 1), for installing a new application with a
service provider and a number of service requesters,
which takes into account the current resources of the
embedded network.
As input, the algorithm takes an embedded net-
work N
emb
, a dedicated service provider node v
s
, a set
of service requesters (the clients) C and their corre-
sponding queries Q, and the underlying data model
of the service provider described in an XML schema
XSD. Its outcome is a subnetwork N
T
emb
of N
emb
that
represents the dissemination tree/path from v
s
to all
clients in C and a set F that consists of the selected
nodes with pre- and post-filter properties. Essentially,
the processing steps of the algorithm can be divided
into three parts:
1. After determining the filter grammar G
F
(see
Section 2.2) by the FilterGrammar in line 1, a
suitable pre-filter node is searched. Doing this,
the ClosestPreFilterNode algorithm (line 2) is
WEBIST2014-InternationalConferenceonWebInformationSystemsandTechnologies
44
Algorithm 1: FilterEnabledDissemination.
Require: N
emb
= (V,E,c,w, p), a service provider v
s
,
set of service requesters C = {v
c
1
,...,v
c
n
}, set
of queries Q related to client’s conditions, data
model represented as XML schema XSD.
Ensure: Tree network N
T
emb
with a set F of dedicated
selected nodes with its filter grammars (pre- and
post-filters).
1: G
F
← FilterGrammar(G,Q);
2: v
pre
← ClosestPreFilterNode(N
emb
,v
s
,C, G
F
);
3: N
T
emb
← DisseminationTree(N
emb
,v
pre
,C);
4: F ← PostFilterPlacement(N
T
emb
,Q,v
pre
,G
F
);
5: extendTreeByPreRoute(V
T
,E
T
,V,E,v
pre
);
6: return {N
T
emb
,F}
called. This algorithm determines one process-
ing node v
pre
that is able to run the filter grammar
G
F
as well as that results in overall positive data
dissemination. More precisely, we are not only
considering the quality of the path to a process-
able node v
pre
in terms of connection and device
class, but also the quality from v
pre
to all service
subscribers.
2. Starting with the determined pre-filter node v
pre
we discover an optimized dissemination tree N
T
emb
from this v
pre
. The DisseminationTree algo-
rithm (line 3) will be called to gather such a
tree. Thereby, relevant service data shall be de-
livered from v
pre
to the service subscribers in a
resource-optimized manner. More precisely, the
data shall be routed via high quality connections,
avoid very constrained embedded devices, and be
shared for as long as possible if there are multi-
client destinations. The latter can be fulfilled if
one or more post-filters can be placed that re-
tain the information of the final client destination
nodes or the next post-filter nodes. Consequently,
the DisseminationTree algorithm finds a dissemi-
nation tree from v
pre
to all clients in C that takes
into account the device class and connection qual-
ity metrics as well as the current processing capa-
bility of the potential post-filter placement.
3. Based on N
T
emb
, suitable nodes are selected for the
post-filter functionality to share service data as
long as possible. The PostFilterPlacement algo-
rithm (line 5) realizes this and provides the routing
information for all filter grammars.
Before the FilterEnabledDissemination algorithm
terminates, we extend N
T
emb
by the involved nodes and
connection that leads from v
s
to v
pre
(line 5).
For better clarification and to get an idea how our
algorithm works we will consider an example in the
next subsection applied on the network shown in Fig-
ure 1.
3.3 Example
We will now consider the embedded network
N
emb
= (V, E, c, w, p) which is shown in Figure 1.
Node 1 is a service provider and nodes 3, 4, and 7
are the service requesters with the following 4 query
conditions as presented in Section 2.2:
Node 3: Q
1
and Q
2
Node 4: Q
3
Node 7: Q
4
Figure 3(b) already shows the filter grammar
G
F
based on this query set after applying the
FilterGrammar procedure. Since the service re-
quester node does not provide us with the opportu-
nity to set up a filter mechanism for clients’ subscrip-
tion requests (p(1) = 0), we have to find an alterna-
tive node for placing a pre-filter. In order to do so, we
have to identify any nodes with processing capabili-
ties within the network that have enough resources to
run G
F
. The procedure ClosestPreFilterNode in Al-
gorithm 1 will identify these nodes (3, 4, 5, 8) check
their resource capabilities. If this results in more than
one node, the node which is closest to the service
provider and yields the best cost function value is se-
lected based on test paths to clients in C using the
Dijkstra algorithm (Dijkstra, 1959) with our metrics.
The outcome would be node 5 that is selected as pre-
filter node (v
pre
) running G
F
.
The next step involves determining a data dissemina-
tion tree that spans v
pre
and client nodes 3, 4, and 7.
In order to do so, we will call the DisseminationTree
procedure. This algorithm is based on the concept of
the Kou-Markowsky-Berman (KMB) algorithm (Kou
et al., 1981) which is a well-known heuristic for the
Steiner Tree problem. Based on our metric and cost
model, respectively, Figure 4 shows the outcome of
the DisseminationTree procedure.
The last major processing step in our dissemination
algorithm involves determining suitable post-filter
nodes to enable a high ratio of shared service data
from service provider to service requesters. Start-
ing with (root) node 5, the PostFilterPlacement pro-
cedure will first select all nodes that contain multi
successor branches. Nodes 5 and 8 are candidates.
Since node 5 already is a dedicated pre-filter node,
we will not consider it further and instead check node
8 directly for processability of a post-filter grammar.
The post-filter grammar is constructed based on the
queries that can be reached from node 8. This is
Filter-enabledBinaryXMLDisseminationinEmbeddedNetworks
45
Class:
Proce:
1
3
3
Class:
Proce:
1
2
4
Class:
Proce:
1
2
5
Class:
Proce:
0
3
7
Class:
Proce:
1
3
8
1
1
1
1
Figure 4: Determined tree based on the DisseminationTree
procedure from v
pre
(node 5) to all clients in C (nodes 3, 4,
and 7).
Envel-
ope
Root Grammar
EV(000)
Header
EV(0)
EV(1)
Body
EV(-)
Envelope Grammar
EV(-)
status
EV(-)
Header Grammar
Body Grammar
Temperature Grammar
scale
EV(0)
value
EV(-)
PS={Q , Q }
3
<21.0
AS={Q , Q }
3
EV(11)
Temp-
erature
>20.5
4
4
EV(1)
Figure 5: Post-filter grammar.
true for the queries Q
3
and Q
4
. Figure 5 shows the
post-filter as based on these queries that is constructed
based on the mechanism presented in Section 2.2.
Since node 8 is a node with processing capabilities we
can successfully install the post-filter on this node.
Before the PostFilterPlacement algorithm is termi-
nated, we are going to update the network’s defined
filters in terms of routing information. Node 5 is set
up with the pre-filter G
F
that is shown in Figure 3(b)
and will receive all messages from node 1. In addi-
tion, node 5 with G
F
contains the associated infor-
mation Q
1
and Q
2
related to node 3, Q
3
related to
node 4, and query Q
4
related to node 7. Based on
the post-filter to be placed on node 8, service data that
matches queries Q
3
and Q
4
shall be forwarded to node
8, which then will send the data only once. Thus, G
F
is updated with this information. In summary, we ob-
tain the following routing information:
• G
F
(at node 5): forwards service data to node 3
when queries Q
1
and/or Q
2
match; forwards ser-
vice data to node 8 when queries Q
3
and/or Q
4
match.
• G
0
F
(at node 8): forwards service data to node 4
when Q
3
matches; forwards service data to node
7 when Q
4
matches
This determined data dissemination tree and the filter
placement is also reflected in Figure 2(b).
4 EVALUATION
So as to organize service data dissemination of each
new applied application and to estimate its influence
in terms of traffic and device capacity usage of real
embedded networks we wrote an embedded network
simulator. The simulator provides us with the oppor-
tunity to load particular network topologies and char-
acteristics as well as service provider and the service
subscribers with their queries. Another alternative is
to setup randomized embedded networks by provid-
ing different kinds of generation parameters: number
of nodes, number of different kinds of device classes,
and the ratio of device classes and connection quality.
Based on such a network, we are able to set up new
applications by selecting particular nodes, which op-
erate a service with the provided service description,
and the client nodes that subscribe service data with
the predefined conditions on the service data. We can
then run our dissemination algorithm for each newly
installed application.
In order to evaluate the effectiveness of the approach
presented in this paper, we randomly generated an
embedded network that has a complexity of 50 nodes
with three device classes. This network setup ini-
tially features a balanced ratio of processable and
non-processable nodes. Its class ratio consists of
5 times device classes 1, 10 times device classes 2,
and 35 times device classes 3. Initially, we uniformly
distributed the connection quality weighting values
with numbers between 0.8 and 1. We sequentially in-
stalled five different kinds of applications. In general,
an application is based on a service provider and dif-
ferent kinds of service requesters (the clients). The
distance (in terms of hop count) and client distribu-
tion to the service provider node is increased with
each new installed application. We start with the first
application, which has two clients; subsequently, the
second has 3 clients, the third has 4 clients, there are
5 clients in the fourth application, and finally the fifth
application has 6 different service requesters. For
each installed application we evaluated the service
data dissemination for two variants: Filter-enabled
dissemination (abbreviated with FD) represents our
filter-enabled dissemination approach and the sepa-
rate and direct dissemination (abbreviated with DD)
WEBIST2014-InternationalConferenceonWebInformationSystemsandTechnologies
46
0
5
10
15
20
25
30
35
App1 App2 App3 App4 App5
0
1
2
3
4
5
Number of Class Occurrence
Number of Post-Filter
Cl1 Cl2 Cl3 Cl1 Cl2 Cl3 Cl1 Cl2 Cl3 Cl1 Cl2 Cl3 Cl1 Cl2 Cl3
FD
DD
Post-Filter
(a)
0
10
20
30
40
50
60
70
80
App1 App2 App3 App4 App5
0
0.2
0.4
0.6
0.8
1
Number of Occurrence
Connection Quality
Total FD
Total DD
Shared FD
Connection Quality DD
Connection Quality FD
(b)
Figure 6: Embedded network with |V | = 50: (a) Count of used classes (Cl1, Cl2, Cl3) in the dissemination path and number
of used post-filters for each application (App1, ..., App5). (b) Number of links of the dissemination path and average value of
connection quality.
reflects the direct, non-filtered service data delivery
(comparable with Figure 2(a)).
Figure 6 shows the evaluation results. Fig-
ure 6(a) depicts the result for each application in
terms of device class occurrences (Cl1=Class 1 nodes,
Cl2=Class 2 nodes, and Cl3=Class 3 nodes) in the
dissemination path of our approach (FD Optimized)
as compared to the simple approach, wherein each
service data is delivered separately (DD Simple). In
other words, we count the occurrence of the device
classes in the determined dissemination path (tree)
that reflects the worst case scenario when a service
message is relevant for all service requesters in the
network. As can be seen for all cases, our approach,
as presented in this paper, results in a lower usage of
class occurrences as compared to the simple service
data distribution variant. This becomes especially
apparent the more complex the application is. Fur-
thermore, the occurrences also show that our deter-
mined dissemination paths always consist of the de-
sirable, relatively small number of constrained nodes
(class 3). For instance, in a worst case distribution
scenario for application five, our dissemination ap-
proach uses the device class 1 sixteen times, class
2 ten times, and the most constrained device class 3
eight times. In total, 34 nodes are involved in the dis-
semination process. In contrast, a simple dissemina-
tion would lead to a device class ratio of class 1 thirty-
five times, class 2 twenty-four times, and class 3
twenty times. In total, this involves 79 nodes. Conse-
quently, our approach results in a better resource us-
age of the nodes in the embedded network since fewer
total nodes are involved in the dissemination tree; the
number of constrained nodes (class 3) is kept as small
as possible.
Figure 6(b) shows the evaluation result in terms of
the number of connection links used and average con-
nection quality. As can be seen, the number of con-
nections used in a dissemination process is smaller
for our approach as compared to the simple variant.
The figure also shows the ratio of the shared connec-
tions of the optimized variant in each application. We
determined the number based on whether each con-
nection between two nodes can reach a pre-filter or a
post-filter node. If so, the number of shared connec-
tions is incremented. The presented numbers show
the effectiveness of our approach, since for each ap-
plication we determine a dissemination tree that con-
sists of a high ratio of shared connections. Figure 6(b)
also shows the average connection quality for each
application and its dissemination based on both our
approach and the simple variant. As can be observed,
the simple dissemination variant loses the average
connection quality faster than our approach. This is
explained by the fact that the simple variant involves a
lot more connection links and potentially causes more
network traffic. The more applications that are in-
stalled in the network, the greater the impact on con-
nection quality will be.
5 RELATED WORK
Finding a suitable pre-filter node outside of the ser-
vice data origin node and the position of post-filters
in a dissemination tree opens the opportunity to share
relevant service data with a number of service sub-
scribers. This leads to a reduction of resources used
within embedded networks in terms of network traf-
fic as well as processing overhead. Similar topics are
addressed by and can be found in Data Stream Man-
agement Systems (DSMSs). DSMSs complement the
traditional Database Management Systems (DBMSs).
Typically, a DBMS handles persistent and random ac-
Filter-enabledBinaryXMLDisseminationinEmbeddedNetworks
47
cessible data and executes volatile queries. Mean-
while, in DSMSs persistent (or long-running) queries
are executed over volatile and sequential data. Exam-
ples of DSMSs include Aurora (Abadi et al., 2003),
Borealis (Abadi et al., 2005), TelegraphCQ (Chan-
drasekaran et al., 2003), and StreamGlobe (Kuntschke
et al., 2005). The main focus of such systems is on the
efficient processing of potentially infinite data streams
against a set of continuous queries. In contrast to
publish/subscribe systems such as XFilter (Altinel
and Franklin, 2000) or YFilter (Diao and Franklin,
2003), continuous queries in DSMSs can be far more
complex than simple filter subscriptions. Some re-
search developed new query languages such as Win-
dowedXQuery (WXQuery) (Kuntschke, 2008) to ex-
tend query operations. In the domain of constrained
embedded networks, however, we presume the pres-
ence of relatively simple data models and have found
that XPath expressions are sufficient to address data
interests and simple constraints by predicates. Other
important work in distributed DSMSs such as Stream-
Globe and Borealis revolves around network-aware
stream processing and operator placement. These
are issues also relevant to constrained embedded net-
works and, similarly, we took them into account
for our approach by positioning the pre-filter and, if
possible, the post-filter mechanism at the embedded
nodes.
Most DSMSs, such as TelegraphCQ for example,
are based on relational data. StreamGlobe, however,
focuses on plain-text XML data streams as well as on
XML-based query languages such as XQuery (Boag
et al., 2007) or the above mentioned WXQuery. Con-
sequently, nodes used for distributed data stream pro-
cessing in systems such as StreamGlobe and Borealis
generally need to be far more powerful than the mi-
crocontrollers for constrained embedded devices that
we aim for in this paper. Our approach for construct-
ing high performance filter mechanisms based on bi-
nary XML techniques enables us to bring DSMS top-
ics to the domain of constrained embedded networks.
In our approach, filter nodes such as pre-filter
or post-filter nodes decide how to best forward ser-
vice messages if there are one or more matches.
The destinations may include service requester nodes
and/or other post-filter nodes. In the literature, this
is called content-based routing or application-level
routing since routing depends on the contents of
data within a message. In that context, we can re-
fer to works such as the combined broadcast and
content-based (CBCB) routing scheme (Carzaniga
and Wolf, 2002), the application layer multicast al-
gorithm (ALMA) (Ge et al., 2006), the usage of XML
Router (Snoeren et al., 2001), and view selection for
stream processing based on XML data (Gupta et al.,
2003). Below, we will concentrate on the latter since
they also involve XML-based data content.
The XML Router approach (Snoeren et al., 2001)
creates an overlay network that is implemented by
multiple XML routers. An XML router is a node
that receives XML packets and forwards a subset of
these XML packets. The XML packets are forwarded
to other routers or the final client node destinations.
Thereby, the output links represent the XPath queries
that describe the portion of the router’s XML stream
that should be sent to the host on that connection
link. XML routers are comparable to our pre- and
post-filter concept. However, additional strategies,
such as reassembling a data packet stream from di-
verse senders provided by the diversity control proto-
col (DCP) or the usage of plain-text XML and XPath
interpreters are not feasible in a resource constrained
embedded environment.
The view selection for stream processing method
is an interesting approach followed in (Gupta et al.,
2003) and (Gupta et al., 2002). The main concept in-
cludes selecting a set of XPath expressions which are
called views. The service data producers evaluate the
views and add the result to the data package in the
form of a header. The advantage is that servers which
keep a local set of queries can evaluate their workload
by inspecting only the values in the header and do not
need to parse the XML document. This leads to a
speed-up of routing decisions. However, this is only
true for cases in which the evaluation in the header is
positive. Otherwise, the complete (plain-text) XML
document needs to be parsed and the query needs to
be evaluated in a conventional way. Again, this is an
obstacle in the constrained embedded environment. In
addition, one of our goals is to achieve seamless pro-
tocol usage and to work with standardized message
representations to support interoperability in a hetero-
geneous network environment. Adding a header to a
message would break this principle and necessitate an
adjustment of communication protocols.
6 CONCLUSION AND FUTURE
WORK
In this paper, we presented an approach to realize
efficient filter-enabled service data dissemination in
constrained embedded networks based on XPath ex-
pressions given by different subscribers/clients. Find-
ing a suitable pre-filter node in an embedded net-
work leads to an early evaluation of relevant ser-
vice messages. By using post-filters in a determined
dissemination tree, we are able to avoid redundant
WEBIST2014-InternationalConferenceonWebInformationSystemsandTechnologies
48
transmissions and share the service data, especially if
there is a multi-query match of different kinds of ser-
vice requesters. The effectiveness in terms of device
class occurrences, connection quality, and number of
shared connections was demonstrated in a simulated
environment based on our embedded network simula-
tor.
Topics for future work include the dynamic update
of client queries and their impact on the dissemination
path as well as on the placed pre- and post filter gram-
mar.
REFERENCES
Abadi, D. J., Ahmad, Y., Balazinska, M., C¸ etintemel, U.,
Cherniack, M., Hwang, J.-H., Lindner, W., Maskey,
A., Rasin, A., Ryvkina, E., Tatbul, N., Xing, Y., and
Zdonik, S. B. (2005). The design of the borealis
stream processing engine. In CIDR, pages 277–289.
Abadi, D. J., Carney, D., C¸ etintemel, U., Cherniack, M.,
Convey, C., Lee, S., Stonebraker, M., Tatbul, N., and
Zdonik, S. B. (2003). Aurora: a new model and ar-
chitecture for data stream management. VLDB J.,
12(2):120–139.
Altinel, M. and Franklin, M. J. (2000). Efficient fil-
tering of xml documents for selective dissemination
of information. In Abbadi, A. E., Brodie, M. L.,
Chakravarthy, S., Dayal, U., Kamel, N., Schlageter,
G., and Whang, K.-Y., editors, VLDB 2000, Proceed-
ings of 26th International Conference on Very Large
Data Bases, September 10-14, 2000, Cairo, Egypt,
pages 53–64. Morgan Kaufmann.
Boag, S., Chamberlin, D. D., Fern
´
andez, M. F., Florescu,
D., Robie, J., and Sim
´
eon, J. (2007). XQuery 1.0: An
XML query language. World Wide Web Consortium,
Recommendation REC-xquery-20070123.
Bournez, C. (2009). Efficient XML interchange
evaluation. W3C working draft, W3C.
http://www.w3.org/TR/2009/WD-exi-evaluation-
20090407.
Carzaniga, A. and Wolf, A. L. (2002). Content-based net-
working: A new communication infrastructure. Lec-
ture Notes in Computer Science, 2538:59–??
Chandrasekaran, S., Cooper, O., Deshpande, A., Franklin,
M. J., Hellerstein, J. M., Hong, W., Krishnamurthy,
S., Madden, S., Raman, V., Reiss, F., and Shah, M. A.
(2003). Telegraphcq: Continuous dataflow processing
for an uncertain world. In CIDR.
Diao, Y. and Franklin, M. J. (2003). High-performance
xml filtering: An overview of yfilter. IEEE Data Eng.
Bull., 26(1):41–48.
Dijkstra, E. W. (1959). A note on two problems in connex-
ion with graphs. Numerische Mathematik, 1:269–271.
Ge, M., Krishnamurthy, S. V., and Faloutsos, M. (2006).
Application versus network layer multicasting in ad
hoc networks: the ALMA routing protocol. Ad Hoc
Networks, 4(2):283–300.
Goldman, O. and Lenkov, D. (2005). XML bi-
nary characterization. World Wide Web Con-
sortium, Note NOTE-xbc-characterization-
20050331. http://www.w3.org/TR/2005/NOTE-
xbc-characterization-20050331.
Gudgin, M., Hadley, M., Mendelsohn, N., Moreau, J.-
J., and Frystyk Nielsen, H. (2003). SOAP version
1.2 part 1: Messaging framework. World Wide
Web Consortium, Recommendation REC-soap12-
part1-20030624.
Gupta, A. K., Halevy, A. Y., and Suciu, D. (2002). View
selection for stream processing. In WebDB, pages 83–
88.
Gupta, K., Suciu, and Halevy (2003). The view selection
problem for XML content based routing. In PODS:
22th ACM SIGACT-SIGMOD-SIGART Symposium on
Principles of Database Systems.
IEEE (2011). IEEE Standard for Local and metropoli-
tan area networks–Part 15.4: Low-Rate Wire-
less Personal Area Networks (LR-WPANs).
http://standards.ieee.org/about/get/802/802.15.html.
K
¨
abisch, S., Kuntschke, R., Heuer, J., and Kosch, H. (2012).
Efficient Filtering of Binary XML in Resource Re-
stricted Embedded Networks. In Proc. of the 8th In-
ternational Conference on Web Information Systems
and Technologies (WEBIST 2012), pages 174–182.
K
¨
abisch, S., Peintner, D., Heuer, J., and Kosch, H. (2011).
Optimized XML-based Web Service Generation for
Service Communication in Restricted Embedded En-
vironments. 16th IEEE International Conference on
Emerging Technologies and Factory Automation.
Kou, Markowsky, and Berman (1981). A fast algorithm for
steiner trees. ACTAINF: Acta Informatica, 15.
Kuntschke, R., Stegmaier, B., Kemper, A., and Reiser, A.
(2005). Streamglobe: Processing and sharing data
streams in grid-based p2p infrastructures. In B
¨
ohm,
K., Jensen, C. S., Haas, L. M., Kersten, M. L., Lar-
son, P.-
˚
A., and Ooi, B. C., editors, VLDB, pages 1259–
1262. ACM.
Kuntschke, R. B. (2008). Network-aware optimiza-
tion in distributed data stream management sys-
tems. PhD thesis, Technical University Mu-
nich. urn:nbn:de:bvb:91-diss-20070806-625762-1-3;
http://d-nb.info/989053113.
Schneider, J. and Kamiya, T. (2011). Efficient XML Inter-
change (EXI) Format 1.0. http://www.w3.org/TR/exi.
W3C Recommendation 10 March 2011.
Snoeren, A. C., Conley, K., and Gifford, D. K. (2001). Mesh
based content routing using XML. In SOSP, pages
160–173.
Filter-enabledBinaryXMLDisseminationinEmbeddedNetworks
49
