A Dynamic Load Balancing Approach Based on the Standard RFID Middleware Architecture

boardpushyUrban and Civil

Dec 8, 2013 (3 years and 9 months ago)

197 views

A Dynamic Load Balancing Approach Based on the Standard RFID Middleware
Architecture
¤
Jae Geol Park,Heung Seok Chae and Eul Seok So
Department of Computer Engineering
Pusan National University
Busan,South Korea
Email:fjgpark,hschae,suk98g@pusan.ac.kr
Abstract
Recently,there has been an increasing interest in
RFID(Radio Frequency Identification) system and RFID
systems have been applied to various applications.Load
balancing is a fundamental technique for providing scal-
ability of systems by moving workload from overloaded
nodes to underloaded nodes.This paper presents a new
load balancing approach by considering the unique features
of RFIDmiddlewares.In contrast to most of the existing ap-
proaches where independencies are assumed among jobs,
applications in RFIDmiddlewares have inter-dependencies
due to RFID readers.Our approach incorporates such
unique characteristics of RFID middlewares.
Keywords:load balancing,RFID middlewares,EPC
Network Architecture
1 Introduction
In recent years,RFID technology has attracted great at-
tention due to technology advancements,heightened secu-
rity concerns and a competitive business environment with
emphasis on cost control and affordable RFIDtag costs [8].
By identifying and tracking each product using RFID tech-
nology,it is possible to collect accurate,immediate infor-
mation about the location and the history of products.RFID
systems have been applied to various applications such as
retail,healthcare,logistics,automotive,food industry,etc.
Wal-Mart,Tesco,Metro together with the Department of
Defense are even demanding the implementation at pallet
level for warehouse and logistics operations.
¤
“This work was supported by the Korea Research Foundation Grant
funded by the Korean Government(MOEHRD)” (The Regional Research
Universities Program/Research Center for Logistics Information Technol-
ogy)
As in conventional distributed systems,load balancing
is also an important feature in RFID middlewares [7].For
example,heavily loaded RFID middlewares may result in
late response or even broken down,which ultimately can
cause untimely behavior of critical business processes.By
applying load balancing techniques to RFID middlewares,
we can resolve such problems.In other words,it is possible
to resolve the problems by redistributing some RFID appli-
cations from overloaded middlewares to other underloaded
ones.
This paper presents a new load balancing approach for
RFID middlewares.Our approach is based on the ALE
(Application Level Events) [4],the standard architecture of
RFIDmiddlewares,proposed by EPCglobal Inc [5].We de-
fine the workload of jobs in RFID middlewares to be char-
acterized by ECSpecs in ALE which are issued by RFIDap-
plications.Hence,workloads of RFID middlewares can be
managed by migrating ECSpecs from heavily loaded mid-
dlewares to lightly loaded ones.We describe basic con-
cepts of RFID middlewares and propose a load balancing
approach especially for RFID middlewares.
2 Overview of RFID Middleware
RFID middleware(formerly called Savant) is a middle-
ware software designed to process the streams of tag com-
ing from one or more reader devices.RFID middleware
performs filtering,aggregation,and counting of tag data,
reducing the volume of data prior to sending to RFID Ap-
plications.These functions are required in order to handle
the extremely large quantities of data that RFIDsystems can
generate through the continuous interrogation of tags.
The ALE(Application Level Events) is proposed to pro-
vide a flexible interface to a standard set of functions which
are expected to be implemented by RFID middlewares.
That is,the ALE defines an interface to a standard set of
IEEE International Conference on e-Business Engineering
0-7695-3003-6/07 $25.00 © 2007 IEEE
DOI 10.1109/ICEBE.2007.17
337
IEEE International Conference on e-Business Engineering
0-7695-3003-6/07 $25.00 © 2007 IEEE
DOI 10.1109/ICEBE.2007.17
337
IEEE International Conference on e-Business Engineering
0-7695-3003-6/07 $25.00 © 2007 IEEE
DOI 10.1109/ICEBE.2007.17
337
accumulation,filtering,and counting operations.
Figure 1 illustrates the role of ALE within an RFID sys-
tem.An RFID application requests an RFID middleware to
collect and transmit EPCs information through ALE inter-
face.In requesting EPCs information,the RFIDapplication
specifies what EPCs relevant to it by ECSpec.Three fields
of ECSpec defines the actual EPCs to be transmitted to an
RFID application.The field readers specifies the readers
from which EPCs information is collected.An RFID ap-
plication can use EPCs information from a part of readers
which are connected to the middleware.The field bound-
aries specifies the time duration the collection and reports
take place.The field reportsSpecs specifies kinds of filters
which are used to exclude EPCs irrelevant to the RFID ap-
plication.
Figure1.Anarchitectureof RFIDmiddlewares
In response to the request,the RFID middleware col-
lects and filters EPCs fromreaders on the basis of the given
ECSpec.The ECReports defines the report produced by
the RFID middleware in response to an ECSpec.In addi-
tion,an RFID middleware can produce continuously multi-
ple ECReports in response to one request of ECSpec.That
is,once an RFID application specifies its relevant EPCs by
defining the readers,duration,and filters,then it can con-
tinuously collect its interesting EPCs from multiple ECRe-
ports
1
.Moreover,in the case of a continuous report,the re-
port destination is explicitly specified in the request.There-
fore,ECReports are transmitted to the location specified in
the ECSpecs regardless of requesting applications.The cur-
rent ALE supports HTTP,TCP,and FILE Notification URI
for specifying the report destination.
The ALE interface uses a logical name for a reader of in-
terest.A logical reader has an abstract name to refer to one
or more physical readers which can be treated as one group.
For example,there may be three RFIDreaders for a loading
dock door in a warehouse.It will be convenient to group
those three physical readers into one logical reader.RFID
1
ALE supports a synchronous request in addition to this asynchronous
request.
Figure 2.Fundamental concepts in ALE
middlewares maintain the mapping information between a
logical reader and its physical readers.Given the logical
readers in ECSpec,an RFID middleware tries to communi-
cate with the corresponding physical readers using the map-
ping information.Figure 2 is a class diagram which illus-
trates the relations among the concepts mentioned.Note the
cardinalities between the concepts.
3 Load Balancing of RFID Middlewares
3.1 Load Balancing by Migrating ECSpecs
The major load of an RFID middleware results from
gathering and filtering of RFID tags information collected
from its connected readers.The number of tags to be pro-
cessed obviously contributes to the load of the middleware.
Therefore,to reduce the load of an middleware,we need
to manage the number of tags which should be processed
by the middleware.As discussed in Section 2,an EC-
Spec issued by an RFID application defines the set of read-
ers and tags which are of interest to the RFID application.
That is,an ECSpec determines the set of readers and con-
sequently the tags information which should be processed
by the RFID middleware.Therefore,we can manage the
load of an RFID middleware by controlling ECSpecs to be
allocated to the middleware.
Let us explain the overall approach to load balancing
for RFID middlewares using an example in Figure 3.Fig-
ure 3 (a) shows two application servers each of which runs
one RFID application,respectively.A total of six ECSpecs
have been allocated to three middlewares.A line between
an RFID application and an ECSpec denotes that the RFID
application issues the ECSpec.
Let us assume that we find out that Middleware1 is
overloaded with the huge amount of tags resulting from
three ECSpecs ECSpec1,ECSpec2 and ECSpec3.To
reduce the load imposed on Middleware1,we need to
move some ECSpecs in Middleware1 to other under-
loaded middleware.For example,we can reduce the load
338
338
338
(a) Before
(b) After
Figure 3.An example for load balancing of
RFID middlewares
by moving ECSpec3 allocated to Middleware1 to other
under-loaded middleware(e.g.Middleware2).Since EC-
Specs directly determine the amount of tags which a mid-
dleware should process,all the load induced by ECSpec3
is nowtransferred fromMiddleware1 to Middleware2 by
migrating ECSpec3(See Figure 3 (b)).
Note that the migration of ECSpec3 causes no change to
RFIDApplication1 that issued ECSpec3 and utilizes the
output of ECSpec3.RFID Application1 even does not
notice that one of its ECSpecs,ECSpec3,has been moved
from the initially allocated middleware,Middleware1,to
other middleware Middleware2.This is the consequence
of the properties of ECSpec.As described in Section 2,each
ECSpec specifies the destination location where the result
of ECSpec is transferred by Notification URI.Therefore,
the result of ECSpec is still routed to the original destina-
tion even after its migration.So,we can safely move some
ECSpecs to other light-loaded middleware without modify-
ing the associated RFID applications.
3.2 Clustering of ECSpecs based on Reader De-
pendencies
As mentioned above,we can migrate some ECSpecs
from a overloaded middleware to some other under-loaded
middleware.In addition,since RFID application has no de-
pendency on the exact RFIDmiddleware,the migration can
take place without considering the associated RFID appli-
cations.Therefore,we can ignore the associated RFID ap-
plications when selecting ECSpecs to be migrated.
In contrast to the independence of RFIDapplications,an
ECSpec has relationship with RFID readers,which should
be considered in migrating ECSpecs.As discussed in Sec-
tion 2,each ECSpec has a specification on the logical read-
ers of interest to the ECSpec.That is,the ECSpec specifies
the logical readers from which it need to used tags infor-
mation.Since an RFID reader is generally connected to
one RFID middleware
2
,the readers that are specified by an
ECSpec need to be also reallocated to other middleware to
which the ECSpec is moved.For example,consider a RFID
systemconfiguration in Figure 4 which shows the deployed
ECSpecs,and the relationships of RFID readers to the EC-
Specs.Let us assume that ECSpec3 has been chosen to be
moved to Middleware2.In that case,the associated read-
ers R4 and R5 should be moved to Middleware2 as well
as ECSpec3 since ECSpec3 needs R4 and R5 and each
reader can be connected to only one RFID middleware.
(a) Before
(b) After
Figure 4.Considering reader dependencies
The relationship of ECSpecs to RFID readers can also
cause another dependency between ECSpecs.For example,
the Reader R5 is shared by two ECSpecs ECSpec3 and
ECSpec4 in Figure 4 (a).Once ECSpec3 is chosen for mi-
gration,then ECSpec4 should be also moved to the same
destination middleware,since ECSpec4 has a dependency
on the reader R5 that is also used by ECSpec3.That is,in-
stead of moving individually an ECSpec,we should move
all the other ECSpecs which share readers with the EC-
Spec.Figure 4 (b) shows the configuration after ECSpec4
as well as ECSpec3 have been moved to Middleware2.
The figure also shows that the associated readers R4 and
2
Recently,we can see some readers which support the connection to
two or more middlewares.However,such multiple connection is not com-
mon yet.
339
339
339
R5 with the moved ECSpecs have been also reallocated to
Middleware2.Due to the space limitation,the detailed de-
scription on the clustering of ECSpecs is intentionally omit-
ted.
4 Related Works
There are many works to provide load balancing appro-
priate for the type of distributed systems [1,11,6,2].Payli
et al.proposed a dynamic load balancing approach to pro-
vide application level load balancing for individual paral-
lel jobs in grid computing environment [10].In network
of workstations environment,there is also work concern-
ing load balancing [12].Agent-based approaches have been
tried to provide load balancing [3,9].
Although there are numerous works on load balancing
strategies,there are limitations in applying them to RFID
middlewares.First,most of them assume independency
among application allocated to middlewares.However,as
discussed in Section 3,RFID applications cannot be re-
distributed without considering the relationship with other
RFID applications.Second,the existing works describe
load balancing strategies in a general manner,which should
be elaborated and adapted into a specific distributed system.
This paper provides a detailed load balancing strategy,par-
ticularly focusing on job selection policy.In addition,the
proposed approach is based on the ALE,the standard in-
terface to RFID middlewares,proposed by EPCglobal,Inc.
Thus,our approach can be applied to RFID middlewares
which conformto the ALE standard.Some ALE-compliant
RFID middlewares provide load balancing.However,their
load balancing approaches are just application of the con-
ventional approaches for distributed systems without con-
sidering the characteristics of RFID middlewares,particu-
larly of ALE standard.
5 Conclusion and Future Work
In contrast to conventional distributed systems,RFID
middlewares have their own characteristics which should be
considered and exploited for more efficient load balancing.
We note that RFIDmiddlewares can be managed by moving
a set of ECSpecs from heavily loaded middleware to other
lightly loaded one.We proposed that a set of ECSpecs be
grouped into the unit of the reallocation by considering the
mapping information among ECSpecs,logical readers,and
physical readers.
We have designed a load balancing system for RFID
middlewares.We will complete the development and ex-
plore its effectiveness by applying it to real RFID middle-
ware systems.In addition,we plan to research on other load
balancing policies specific to RFID middlewares including
job transfer policy and location policy.
References
[1] M.S.Al-Amri and R.E.Ahmed.New job selection
and location policies for load-distributing algorithms.
Int.Journal of Network Management,12(3):165–178,
2002.
[2] J.Balasubramanian,D.C.Schmidt,L.W.Dowdy,and
O.Othman.Evaluating the performance of middle-
ware load balancing strategies.In EDOC,pages 135–
146,2004.
[3] J.Cao,D.P.Spooner,S.A.Jarvis,S.Saini,and
G.R.Nudd.Agent-based grid load balancing us-
ing performance-driven task scheduling.In IPDPS,
page 49,2003.
[4] EPCglobal Inc.The application level events (ale)
specification,version 1.0.
[5] EPCglobal Inc.http://www.epcglobalinc.com.
[6] P.Krueger and N.G.Shivaratri.Adaptive location
policies for global scheduling.IEEE Trans.Software
Eng.,20(6):432–444,1994.
[7] S.Leaver.Evaluating RFID middleware.Technical
report,Forrester Research,Inc.,2004.
[8] E.Michielsen.RFID middleware market competition
heats up.RFIDInternational Newsletter,2(16),2004.
[9] C.C.Myint and K.M.L.Tun.A framework of us-
ing mobile agent to achieve efficient load balancing in
cluster.In In:Proc.of APSITT,pages 66–70,2005.
[10] R.Payli,E.Yilmaz,A.Ecer,H.Akay,and S.Chien.
A dynamic load balancing tool for grid computing.In
In:Proc.of Parallel CFD,2004.
[11] N.G.Shivaratri,P.Krueger,and M.Singhal.Load dis-
tributing for locally distributed systems.IEEE Com-
puter,25(12):33–44,1992.
[12] M.J.Zaki,W.Li,and S.Parthasarathy.Customized
dynamic load balancing for a network of workstations.
J.Parallel Distrib.Comput.,43(2):156–162,1997.
340
340
340