Opportunistic Channels:Mobility-aware Event
Delivery
Yuan Chen1,Karsten Schwan1,and Dong Zhou2
1College of Computing,Georgia Institute of Technology,Atlanta,GA30332,USA
{yuanchen,schwan}@https://www.doczj.com/doc/4312454092.html,
2DoCoMo USA Labs,San Jose,CA95110,USA
zhou@https://www.doczj.com/doc/4312454092.html,
Abstract.The delivery of data in pervasive systems has to deal with
end host mobility.One problem is how to create appropriate,application-
level data provisioning topologies,termed data brokers,to best match
underlying network connectivity,end user locations,and the locales of
their network access.Another problem is how to balance workloads
in such overlay networks,in response to mobility and to changes in
available processing and communication resources.This paper improves
the performance of data provisioning by dynamically changing broker
topologies and end users’assignments to brokers.Speci?cally,using pub-
lish/subscribe as a communication paradigm,a new abstraction,termed
an opportunistic event channel,enables dynamic broker creation,dele-
tion,and movement.Experimental and simulation results demonstrate
the ability of opportunistic channels to optimize event delivery and pro-
cessing when end users move across di?erent network access points.The
technique is to‘opportunistically’follow network-level hando?s across
network access points with application-level hando?s of a user’s broker
functionality to a new,‘closer’broker.The potential load imbalances
across brokers caused by such hando?s are also addressed.
Opportunistic channels are realized with the JECho event infrastructure.
Performance advantages attained from their use can be substantial,with
the cost of sending a message from a publisher to a mobile subscriber
improved by up to50%.Load balancing improves event delivery even for
moderate numbers of event subscribers.
1Introduction
Publish/subscribe is a widely used paradigm for interconnecting applications in distributed environments.It provides anonymous,inherently asynchronous group communication,where event providers and consumers interact via event brokers,as illustrated in Figure1.Subscription means that an event consumer declares its interest in receiving certain events,using some predicate or more generally,in content-based subscription,a?lter/conversion function de?ned on event contents(e.g.,see the ECho and JECho infrastructures developed in our re-search[9,38]).An event provider generates and publishes events,where message brokers are responsible for collecting event subscriptions and routing events from
publishers to interested consumers.Each provider/consumer connects to one of the brokers,where broker networks can be organized in multiple ways,rang-ing from single central broker(Elvin[31]),to hierarchical topologies(JEDI[7], Gryphon[24]),to general graphs(SIENA[3],READY[11]).Once a topology of brokers has been de?ned,appropriate routing paths must be established to ensure the correct and e?cient delivery of events to all interested consumers.
provider
provider consumer consumer
consumer
Fig.1.Publish/Subscribe Communication Model The issue addressed in this paper is that publish/subscribe systems’broker infrastructures or more generally,overlay networks[20]used in distributed appli-cations do not take into account the unique needs of mobile systems.Problems include(1)which interconnection topology may currently best suit the underly-ing mobile system’s communication paths[26]and how to dynamically construct such a topology,(2)selecting the numbers and capabilities of brokers to match the mobile platform’s con?guration,and(3)dynamic changes in(1)and(2) to match changes in mobile node locations[35],con?gurations,and resources. For publish/subscribe systems,this implies the need to use dynamically varying topologies with a changing number of brokers,and it requires runtime changes to providers’and consumers’assignments to brokers.A speci?c example is one in which a location change by an event provider or consumer results in a con-sequent change of the network access point used by the underlying Mobile IP protocol[25,17].This can lead to ine?ciencies in broker communications,as depicted in Figure2,where a mobile client’s crossing of a network boundary results in a circuitous path from publisher to home broker to event subscriber.
Opportunistic channels is an event channel concept that addresses the mobil-ity of event producers,consumers,and of the dynamic changes in the pervasive systems where they operate.The following attributes of opportunistic channels di?erentiate them from previous work:
–Network and location awareness.The event broker associated with an op-portunistic channel is aware of the underlying network topology used for transporting events from providers to consumers.It is also aware of the re-spective locations of both.
–Dynamic broker adaptation.Brokers can be created,deleted,and moved at runtime,resulting in the ability of opportunistic channels to react to changes
Fig.2.Ine?cient Event Delivery after Subscriber Migration in end user locations and also to secondary e?ects of such changes,such as broker overloads due to changes in client/broker assignments.
The algorithm used for dynamic broker deployment and movement is one that evaluates the performance opportunities presented by current network connec-tivities and broker loads/node capabilities.Speci?cally,it attempts to optimize the event delivery path when a mobile user moves out of range of its old net-work access point by‘opportunistically’following the network level hando?[25, 35]with an application-level hando?of the user’s brokering functionality to a broker that is‘closer’to the new client location,hence the term‘opportunis-tic’event channel.The potential load imbalances across brokers caused by such middleware-level hando?s are addressed by dynamic broker creation.In e?ect, we aim to dynamically construct portions of the event dispatch trees used in static broker architectures[24].
Opportunistic channels are realized with the JECho peer-to-peer pub/sub infrastructure[38].A unique extension of JECho is that its current peer-based broker infrastructure(i.e.,each peer runs its own broker)is enhanced with dy-namically created‘third party’brokers,which can run on machines and/or in address spaces not used by event publishers or subscribers[38].Moreover,gener-alizing the capabilities of other event systems,brokers perform tasks in addition to the event routing permitted by other systems[24],using subscriber-provided functions,termed event modulators[37].The intent is to address the severe re-source limitations existing in many mobile and embedded systems,by permitting event consumers to deploy application-speci?c functions that manipulate event content into event sources and/or brokers,so as to precisely meet their current needs,and to avoid needless data transfers[38].Finally,since JECho is con-structed with Java,the realization of opportunistic channels presented in this paper has some limitations,including the need for a reasonably sized Java foot-prints on the target mobile nodes[38].Opportunistic channels,however,also
bene?t from certain JECho functionality,such as the relative ease with which third party brokers may be created,deployed,and migrated.
The performance advantages attained from the use of opportunistic channels can be substantial.For instance,simulation results indicate that the cost of sending a message from a?xed publisher to a mobile subscriber can be improved by up to50%.Experimental results show that the end-to-end latency can be improved by20%with microbenchmarks,when a mobile devices uses a802.11a-based wireless network,where access points are connected via a high performance network backbone.Furthermore,broker load balancing can deliver events at two times the speed of a non-load balancing solution when the number of subscribers reaches16,as demonstrated with microbenchmarks.Finally,even for a moderate number of video players(i.e.,8players),load balancing and dynamic broker creation permits video playout to be improved by a factor of3.
In the remainder of this paper,Section2brie?y reviews the JECho event system,clarifying the basic software architecture of opportunistic channels.In Section3,opportunistic channels are used in a mobile environment,including experimental measurements with campus wireless network.Section4describes load balancing in opportunistic channels.Related work is discussed in Section 5,and conclusions and future work appear in Section6.
2System Architecture
2.1Overview of JECho
JECho implements a publish/subscribe communication paradigm,providing in-teractive services to distributed,concurrently executing components via event channels.JECho’s e?cient implementation enables it to move events at rates higher than other Java-based event system implementations[38,37].In addi-tion,using JECho’s modulator concept[38,37],individual event subscribers can dynamically tailor event?ows to their own needs,and adapt to runtime changes in component behaviors and needs and/or changes in platform resources.JE-Cho’s implementation is in pure Java,its group-cast communication layer is based on Java Sockets,and it runs with both standard and embedded JVMs.A modulator is a Java object that executes in a source’s or broker’s address space, on behalf of some client.The basic communication model in JECho is shown in Figure3.
provider consumer
Fig.3.JECho Communication Model
Client-de?ned customization in modulators,including event conversion or transformation,is performed prior to delivering the event to the consumer.Event conversion may reduce their sizes and hence reduce network tra?c.Modulators
can also be used for o?oading computation from constrained mobile devices.An example is a modulator that pre-converts events to the forms needed by speci?c client’s graphical displays,thereby eliminating the costs of data conversion in the client.In fact,it is sometimes impossible to display data appropriately without performing such conversions,as with rendering OpenGL-based graphical data on a PalmTop or when applying server-resident business rules to data prior to its display on cellphones[6].Other reasons for such conversion include the delivery of data with certain Quality of Service or to conserve power on battery-limited handheld devices[28].
2.2Broker Adaptation Architecture
The notion of opportunistic channel(OC)generalizes upon JECho’s modulators in three ways.First,OC modulators can run in consumers,intermediate brokers, or in providers.Second,brokers may be created dynamically.Third,for load balancing and migration,each broker is associated with a Java object called an adaptor.The idea is for a modulator and adaptor to adapt event dispatching to the dynamic needs of the clients,as shown in Figure4.
Broker Information Server
broker
Fig.4.Opportunistic Channel Architecture
The adaptation framework constructed with brokers and adaptors utilizes a resource monitoring service(RMS),which collects,aggregates,processes,and delivers data about the current execution environment.The RMS is able to mon-itor any nodes in the mobile system,using daemon processes connected via their own communication links.Monitoring data is captured from objects(including devices)at these nodes,and it is distributed to RMS clients via a customizable push-based interface.As with other monitoring systems[12],such data is con-tained in events,with current events containing information about CPU loads, memory availability,hando?actions occurring at the network level(to capture client mobility),and application-level communication latency and bandwidth
experienced between RMS clients.RMS clients can selectively register their in-terests in the data being captured,by providing a?lter function installed at the source of a monitoring data stream.For example,a broker may provide a func-tion that evaluates captured data and then provides it with a noti?cation event only when the network topology changes and when the broker’s own current CPU load exceeds80%.Our current prototype implementation uses a central RMS server.
Another component of the architecture is a broker information server,which maintains information about current brokers,including their names,certain at-tributes(e.g.,CPU and memory loads,IP addresses,network-related param-eters,etc.)and their interconnections.A broker can get these information by sending a query to the broker information server.
A sample adaptor used by a broker is shown below.Its method‘subscribe()’registers the broker with the RMS and speci?es its interest in certain monitoring data,using the object‘?lter’.Adaptation code is implemented in the event han-dler method‘push()’,which is invoked whenever an interesting event is received. Using adaptors,RMS,and?lters,system developers can create potentially com-plex adaptation policies[34].
public class MyAdaptor implements BrokerAdaptor{
//join the resource channel
public void subscribe(){
registerToRMS("CPU_Load",filter);
}
public void push(Object e){
//adaptation code
}
In summary,the adaptation model used for realization of opportunistic chan-nels is seamlessly integrated with the JECho publish/subscribe system.Adapta-tion involves broker registration with a resource monitoring service and the use of potentially broker-speci?c adaptors.The adaptor can perform simple tasks like recording certain resource changes,to complex adaptations like changing the topology of broker interconnection.
3Opportunistic Channels in Mobile Environments
While the concept of opportunistic channels may be applied to a wide variety of wireless networks,this paper assumes that there exists a reliable underly-ing network with guaranteed connectivity,where links are subject to dynamic variations in available communication bandwidth.In the presence of end user mobility,connectivity is guaranteed by Mobile IP,which performs network-level hando?actions[25,17,35].The resulting transparent mobility exists in both de-ployed and experimental systems,including Berkeley’s BARWAN and CMU’s wireless Andrew systems[20,15],which provide such support within a network (horizontal hando?)and across heterogeneous networks(vertical hando?)across
substantial geographical areas.For simplicity,in the remainder of this paper,we ignore any transient disconnection,which may happen during mobile suppliers’or consumers’migration.Furthermore,brokers are pre-created onto all nodes of the underlying network,thereby ignoring the costs of dynamic broker creation.
3.1Mobility Adaptation
The network topologies of mobile systems change in response to mobile users’movements.When using base stations,this implies that the router(network access point)to which a mobile host connects may change when the mobile host moves across network boundaries.In Figure5,M is a mobile host whose home network is network-2.M receives events from a source residing in network-1.There are two brokers in the system:broker-2in network-2and broker-3in network-3.When M?rst subscribed to the event channel,it speci?ed broker-2as its broker and placed its modulator into broker-2.The event was delivered from the source to broker-2through the network and?nally reached M.The resulting delivery path was network-1(source)→network-2(broker-2→M).At some point in time,M moved from network-2to network-3.If M still used broker-2,the result-ing delivery path would be network-1(source)→network-2(broker-2)→network-3(M).However,if M changed its broker from broker-2to broker-3,then the event delivery path would become network-1(source)→network-3(broker-3→M), which is much shorter.
Fig.5.Event Delivery in a Mobile Environment In most cases,using the old,longer dispatching path(i.e.,triangle delivery) results in higher event delivery latency and in the consumption of additional, often scarce network capacity.A solution is to change event delivery paths along with changes in underlying network message delivery.In particular,when a mo-bile consumer moves out of range of its previous network and into the range of a new one,the mobile host should connect to the broker that is‘closest’to the new network access point.This is the basic idea of opportunistic channels(OC):to ‘opportunistically’move broker activities as end users move.Speci?cally,using the adaptation mechanism described in Section2.2,a simple adaptor transpar-ently relocates the mobile client’s modulator to the new broker in response to a
network-level hando?,the latter detected via online monitoring.The basic steps performed by this adaptor are:
1.search for an alternative broker via the broker information server;
2.send a hando?request to the remote broker;and
3.if the remote broker agrees to accept the request,then execute the broker-
level hando?protocol.
The modulator hando?protocol,described in detail in Section3.2and imple-mented by the adaptor,ensures that the dynamic hando?of a modulator is performed without losing or duplicating events.The broker information server lists the available brokers,including their names and certain attributes(e.g., CPU and memory loads,IP addresses,network-related parameters,etc.).The current implementation uses a central broker information server.More detail on scalable directory services appears in[2].
Two di?erent adaptors are implemented and evaluated in this research.The ‘simple’adaptor always performs modulator hando?,with the intent of using the broker that is closest to the current location of the mobile client.A more realistic adaptor takes into account additional factors.First,it may also consider the new broker’s load.Second,it evaluates the network path to the new broker. This is important because the latency from network-1to network-3may actually be larger than the latency of network-1to network-2plus network-2to network-3 shown in Figure5.Speci?cally,the‘complex’adaptor evaluated next compares the old path with the potential new path when network hando?is detected.It schedules the modulator hando?only when the new path is shorter than the old one.In the current implementation,latency is measured by sending a ping’message between two nodes and using one-half of the round-trip time as an approximation.The source of events can be achieved from broker information server.
3.2Modulator Hando?
Modulator hando?is illustrated in Figures6.Our algorithm guarantees correct-ness properties that include(1)in order event delivery,(2)no lost or duplicate events,and(3)consistent modulator state in the presence of migration:
1.The source broker initiates a hando?by sending a HANDOFF request to
the destination broker.
2.Upon receiving the hando?request,the destination broker adds the mobile
client to its consumer list and sends an ACK to the source broker.
3.After receiving the ACK from the destination broker,the source broker sends
a DETOUR request,which includes the name of the destination broker,to
all of the event providers.
4.Upon receiving the DETOUR request,each provider atomically replaces the
source broker from its consumer list with the destination broker;it then sends an ACK to the source broker and the destination broker.
5.The source broker receives events from each provider,applies the client’s
modulator to these event,and forwards them to the client until it receives the ACK from the provider.
6.The destination broker bu?ers events from each provider after it receives the ACK.
7.After ACKs from all providers are received by the source broker,it removes the client from its consumer list and sends a HANDOFF along with the current modulator to the destination broker.
8.Upon receiving the HANDOFF,the destination broker applies the modulator received from the source broker to the bu?ered events and starts forwarding events to the client.
consumer
provider Location1
basestation2Location2Location1
Fig.6.Modulator Relocation
This protocol ensures that no events are lost or duplicated,since the old broker processes all events from a provider before receiving the provider’s ACK mark,while the new broker processes only those events after the ACK message.The order of events from the same provider is maintained,since both the old broker and new broker process the events in the order delivered by the provider.Since modulators are stateful and since event processing may change their states,the protocol described above also ensures the consistency of modulator state.Finally,the algorithm is non-disruptive,in that the relocation procedure does not directly a?ect the other consumers of the providers handled by a certain broker.This is because it requires neither the providers nor the brokers to temporarily stop event delivery.
3.3Experimental Evaluation
This section establishes the basic performance of opportunistic channels,and it demonstrates that opportunistic channels can improve the performance of event delivery in mobile environments.
Relocation Overhead When measured on a cluster of 300Mhz Pentium II Linux PCs connected with dual 100Mbps fast Ethernet,the times required to complete a relocation for a varying number of suppliers (see Figure 7)demon-strate that relocation cost increases linearly with the number of event suppliers.This is explained by the underlying peer-to-peer communications occurring in
JECho,where brokers are integrated with event suppliers unless otherwise spec-i?ed.The intent,of course,is to separate the broker infrastructure from the remainder of the JECho system,so that solutions like opportunistic channels can specialize their broker behavior for the target environments being addressed.Interestingly,even for the simple peer-to-peer interconnections,total relocation cost remains small even for a moderate number of suppliers (i.e.,16suppliers).In these measurements,each supplier is located on a di?erent machine,and the event consumer moves across two Linux machines in the same cluster.050
100
150200250
0246810121416
R e l o c a t i o n C o s t (m s )Number of Suppliers Relocation Cost
Fig.7.Relocation Overhead
Simulation Results For the purpose of simulation,a 100node network is generated using the BRITE (Internet topology generator developed at Boston University [21]).Nodes are assumed to be geographically distributed,and each link is assigned a cost corresponding to the distance between the nodes it con-nects.The cost of sending a message between any pair of nodes is the cost of the shortest path between the nodes,computed with Dijkstra’s algorithm.There are multiple ?xed sources and one mobile host in the network,all located on net-work nodes.Brokers also reside on these nodes.In the simulation,the mobile host stays on the same node for some varying time period,which is generated from an exponential distribution.When the period expires,the mobile host moves to the ‘next’location,according to the mobility model used.In the simulation,four mobility models characterize di?erent user behaviors.The random model chooses the next location randomly.The traveling salesman model moves to one of its neighbors,which is randomly chosen.The pop-up model is similar to the traveling salesman one,except that it sometimes roams to a randomly chosen remote node.The ?xed model has two ?xed locations which the mobile host visits more frequently than others.The mobile host moves between these two locations most of the time.Occasionally,the mobile host roams to a di?erent node.More detail about the client behavior models used in our experimentation appears in [30].
The simulation uses a mobility-to-communication ratio to measure the mo-bile host’s movement speed.This ratio captures the rate at which a mobile host changes its location to the rate at which it receives messages.The higher the mobility-to-communication ratio,the faster the mobile host’s movement.For each relocation,adaptation cost is measured by multiplying the number of messages used by the relocation protocol by the distance between suppliers,consumer,source brokers or destination broker.
The ?rst simulation uses one ?xed source and a single mobile client.We compare the performance of using a ‘normal’channel,a ‘simple’opportunistic
Random
0.000010.000050.00010.00050.0010.0050.01
1000
2000
3000
4000
5000
0.000010.000050.00010.00050.0010.0050.01
Pop-up
0.000010.000050.00010.00050.0010.005
0.01
Fixed
200
400
600
800
1000
1200
1400
1600
1800
2000
0.000010.000050.00010.00050.0010.0050.01
X Y
Fig.8.Simulation Results for Single Supplier
channel(oc1)and a‘complex’opportunistic channel(oc2).oc1uses a simple adaptor,which always performs a modulator hando?when the mobile client moves to a new location.oc2’s adaptor compares the distances of old path and new path when the mobile client moves to a new location.If the new path is shorter,the adaptor relocates the modulator to the new broker.Otherwise, the old broker is used.Figure8shows the average cost per message of three channels,respectively,using the random model,traveling salesman and?xed model.In the?gure,the X-axis indicates the mobility/communication ratio,and the Y-axis indicates the average cost of receiving a message.The results show that opportunistic channels achieve better performance than normal ones,even for random walks with very high mobility.In all cases,complex opportunistic channels(oc2)perform better than simple ones(oc1).In general,when the mobility/communication ratio increases,relocation cost and hence,average cost, increases accordingly.
The second simulation involves?ve?xed sources and one mobile client.The results are shown in Figure9.They demonstrate that the opportunistic channel with the complex adaptor has the best performance,and that even the simple adaptor still outperforms non-opportunistic channels.
End-to-End Delay To substantiate our claims,experiments are performed with a microbenchmark used in three di?erent locations on the Georgia Tech campus,as shown in Figure10.The achieved end-to-end delays using oppor-tunistic vs.normal channels are compared,with an event publisher running on a SUN Ultra30machine(H1)at subnet1.An IPAQ is the event consumer,and it accesses the network via802.11a access points(802.11a devices o?er a maximum of56Mb bandwidth).Initially,the mobile client runs in subnet2and connects to the event broker running on H2,with network connectivity provided via a wire-
0.000010.000050.00010.00050.0010.0050.01
Fig.9.Simulation Results for Multiple Suppliers
less network access point attached to subnet2.When the mobile client moves to another location(subnet3),with non-adaptive channels,the mobile client con-tinues to connect to the old broker(H2),whereas the opportunistic channel uses a local broker(H3).With opportunistic channels,the mobile client and its bro-ker have access to the same local network(subnet2),hence resulting in a shorter path between them.The experiment is performed at midnight in order to avoid high levels of variation in network usage.The results presented here are the av-erage performance achieved over50,000events.The standard deviation for the measurements is less than10%of the average latency.
subnet2
Fig.10.Experiment Con?guration for End-to-End Delay The experiment measures the end-to-end delay with data of size25KB in two di?erent locations,as shown in Figure11.Since the opportunistic channel can always connect the mobile client to a local broker,the latency is nearly iden-tical after moving to new location.In comparison,with the non-opportunistic channel,latency increases by more than20%after migration.This is due to the fact that event delivery for a non-opportunistic channel follows the path
subnet1→subnet2→subnet3,whereas an opportunistic channel directly delivers the event from subnet1to subnet3.The ?gure also shows a breakdown of the costs involved in event latency.The portions labeled ‘wired’represent the la-tencies from the event publisher to the wireless access points.The ‘wireless’times are the delays in transmitting data from access point to mobile client.This breakdown is useful for understanding the performance di?erences between non-adaptive and opportunistic https://www.doczj.com/doc/4312454092.html,ly,in all of these scenarios,the wireless times are nearly identical,since all network accesses use the same 802.11a devices.The di?erence is the time from the publisher to the access point,that is,the ‘wired’times.In this scenario,therefore,the opportunistic channel optimizes ‘wired’times.
05
10
152025
30
l a t e n c y (m s )Fig.11.Microbenchmark End-to-End Delay
Clearly,even in realistic scenarios within a campus like Georgia Tech,per-formance advantages can be derived from using opportunistic channels.The numeric bene?ts attained here are somewhat small,however,because the domi-nant factor in the scenario studied is the communication between broker (access point)and the mobile client.Even with 802.11a,this wireless communication is more than 2times slower than the wired communications.This will not be the case for wide area mobility or in adhoc wireless networks,with potentially large delays between the publisher and the broker.Actually,in ad-hoc wireless networks,the delay between a publisher and a broker should be comparable to the delay between a broker to a mobile client,since both communications use wireless links.Finally,even the relatively modest 20%improvement in latency shown here can be important to applications that require real-time response.Ex-amples include web portals [27]and the interactive virtual workbench for remote machine monitoring and control [6]evaluated in our previous research.4Load Balancing in Opportunistic Channels
Mobile hosts are known to be limited in computing power,storage,display,net-work bandwidth,etc.One approach to dealing with such limitations is per client service di?erentiation,that is,to customize service delivery to the needs of indi-vidual clients.JECho enables such customization by deploying modulators into event sources and/or brokers,where modulators range from simple event ?lters
to complex time-consuming event transcoding engines.With consumers and sup-pliers joining and leaving dynamically,broker loads are also subject to runtime variation,including overloads caused by modulator migration in opportunistic channels.One reason is the arrival of a large numbers of local users,as when many mobile units converge,such as during meetings.Another reason is the use of complex modulators by‘thin’clients,such as modulators that implement the
?exible data transcoding required by such clients
[38].
Fig.12.Modulator Relocation for Load Balancing
4.1Adaptations for Load Balancing
Opportunistic channels use a load balancing adaptor that permits a broker to distribute its load(i.e.,modulator execution)to a less heavily loaded broker. This adaptor works as follows.The broker monitors the total execution time of all modulators in its address space,using broker-resident timers maintained by the resource monitoring service(RMS)described in Section2.When the broker’s modulator execution time exceeds some limit,which can be speci?ed when the broker is started,it will try to?nd a less loaded broker and relocate some of its load(i.e.,modulators)to that broker.In the current implementation, adaptors simply contact a directory service to identify a lightly loaded broker. The directory service?nds a lightly loaded broker as follows.First,it has infor-mation about each broker’s location,current load,status indicating whether it can receive the load balancing request and network distances between all pairs of brokers.Second,using the RMS,it collects information that includes remote brokers’CPU loads,modulator execution times,and total execution times.The frequency with which such monitoring information is exchanged is also set when the broker is started.Third,given this information,the directory executes the following,simple algorithm to identify a suitable broker:
-check all available brokers in the order of the distance from the request-ing broker;
-if a less loaded broker is found,stop the search and return the broker’s id;
-if no such broker exists,terminate with failure.
The broker initiating the request contacts the target broker and executes the relocation protocol described in Section 3.2.To reduce the frequency with which relocations are performed,our current implementation simply relocates half of the modulators from the overloaded broker to the target broker.The result of these actions is depicted in Figure 12.
4.2Experimental Evaluation
Benchmarks Figure 13compares the performance of event distribution for op-portunistic vs.normal channels.The events being passed are arrays of 100?oats.A for()loop is used to emulate the behavior of complex modulator runtimes.When brokers and consumers reside on di?erent machines,with opportunistic channels,one intermediate broker is added after 6consumers have joined the channel,and a second broker is added after 14consumers have joined.In com-parison,with the non-opportunistic channel,all consumers use the same broker for the entire duration of the experiment.The results clearly establish the im-portance of dynamic load distribution across brokers.0.51
1.5
2
2.5
3
3.5
0246810121416
T i m e p e r e v e n t (m s )Number of consumers Without OC With OC
Fig.13.Load Balancing Microbenchmark Results
Performance of a Video Player A more realistic experiment is one that measures the frame rate of a video player,varying the number of players and comparing the results with a single broker vs.dynamically ‘split’,multiple bro-kers.In the experiment,a video server runs on a Linux machine and multiple players each running on a di?erent Linux machine receive video stream from the server.The results are shown in Figure 14.For 8players,opportunistic channel permits the frame rate to be improved by a factor of 3.
5
Related Work 5.1Publish/subscribe Systems
Most current publish/subscribe systems permit subscribers to specify their in-terests,storing these interests at providers or intermediate brokers.Typical in-terest expressions result in predicate-based event ?ltering,sometimes also en-abling limited forms of event transformation.However,most such subscription
05
101520
12345678
F r a m e s p e r s e r o n d (f p s )Number of consumers Without OC With OC Fig.14.Frame Rate Comparison
are stateless and therefore,do not support the general event processing needed for the complex data conversions occurring in multimedia,business,or scienti?c applications.In addition,a client’s subscriptions cannot be changed at runtime.JECho and opportunistic channels generalize upon these systems’capabilities in our support for dynamic broker creation,load balancing across brokers and most importantly,in our ability to make broker behavior platform-aware,by dynamically monitoring and adapting to changes in platform capabilities and client behaviors.
There has been substantial research on interconnection topologies,event rout-ing,and event matching (to eliminate redundant event transmission)in order to scale publish/subscribe systems to large numbers of publishers,subscribers,and events.Typically addressing wide area networks (e.g.,Gryphon [24],SIENA [3],JEDI [7],READY [11]),The main goal in such work is to ‘send’an event only to-wards brokers with consumers that are currently interested in that event,and to use shortest such path available.In comparison to our work,?xed topologies are used for broker interconnection,and there is no dynamic support for changing topologies or broker locations.More speci?cally,JEDI o?ers moveout and movein operations to change its dispatching system somehow,but the applications have to explicitly call those operations [8].Elvin supports disconnection and recon-nection in a mobile environment by using proxies,but the mobile client must connect to the same proxy even after it moves to a new location [33].Neither system provides general adaptation support for dynamic recon?guration and re-deployment of the event system itself.[16]discusses general ideas about how to adapt a publish/subscribe system to mobile environments,but does not describe an implementation.Solar [5]provides a ?exible and scalable data-fusion infras-tructure for delivering context information in ubiquitous computing systems.Its concept of operator is similar to modulators in our systems,but modulator in JECho can perform more general operations.Furthermore,Solar focuses on how to delivery context information e?ciently,whereas JECho is a general event system.
5.2Mobile IP and its Location Management
Optimization of location management for Mobile IP [19]is similar to the way in which opportunistic channels https://www.doczj.com/doc/4312454092.html,ly,both attempt to optimize event routing paths after a mobile host changes its location,by changing the ‘trian-gle’delivery to direct delivery.Multiple hando?algorithms have been proposed for Mobile IP [19,13],all of which require location registration with clients’
home agent or a regional agent.In comparison,opportunistic channels use di-rectory services in place of home or regional agent,which implies that hando?performance depends on that of the directory service.Furthermore,modula-tor hando?is somewhat more complex than link hando?,especially for stateful modulators.Finally,since opportunistic channels are realized on top of Mobile IP,its performance bene?ts and overheads depend on the underlying Mobile IP implementation.
5.3Adaptations in Mobile Environments
Several network-level e?orts address mobility issues.As already mentioned,Mo-bile IP(e.g.,Mobile IPv6[25,17])enables a mobile user to stay connected when moving across network boundaries without changing its IP address.Berkeley’s BARWAN project[20]provides seamless roaming across heterogeneous networks. Furthermore,BARWAN enables data forms to be changed to suit end system or wireless network limitations,by permitting application-level,type-speci?c data transformation and data compression[20].JECho o?ers the same functionality, but in addition,provides dynamic support for handler partitioning[37].Zhao, Castelluccia and Baker[36]describe a general-purpose mechanism at the network level,which supports multiple packet delivery methods and multiple network in-terfaces,where the system adaptively selects the most appropriate method and interface.Similarly,the CMU Monarch project aims to enable adaptive mobile host communications,to make the most e?cient use of the network connectiv-ity available to the mobile host at any time[18].The adaptors used in oppor-tunistic channels would interact with such network-level mechanisms.Resilient Overlay Networks[1]optimize application-speci?c routing metrics,by monitor-ing the functioning and quality of network paths.Opportunistic channels could implement the same mechanisms,if appropriate.Finally,there are several TCP enhancements for wireless networks[4].
Our work focuses on middleware-level and application-level adaptations,and it can bene?t from the network level research mentioned above.Since such adap-tations require information about network-level changes,we can also bene?t from the substantial ongoing work on real-time network monitoring,including the work performed in the Monarch[18]and net100[22]projects.
At application-level,the Odyssey projects extends the Unix System call in-terface to support?exible application-aware adaptations[23],as also done in our own work addressing interactive applications[29].The system monitors resource levels,noti?es applications of relevant changes,and enforces resource allocation decision.Each application independently decides how best to adapt when noti-?ed.This is similar to JECho’s adaptations where an application can be noti?ed of resource changes and responds to such changes according to its adaptation strategy de?ned in a modulator.
Some adaptations based on application semantics can be provided only at application level.For example,datatype-speci?c data transformation and data compression must depend on the application.Similarly,HRL’s Intelligent in-formation dissemination services use bandwidth-aware?ltering to adapt infor-
mation streams to resource bandwidth availability[32].Our own previous work with JECho has addressed these problems by supporting general application-level adaptations via the dynamic deployment and partitioning of event modula-tors[38].Opportunistic channels,as well as our related research on coordinating network-with application-level adaptations[14]complement these e?orts,by providing models and mechanisms for linking network-or broker-level changes in resource availability with suitable application-and middleware-level adapta-tions.
6Conclusions and Future Work
This paper describes how a content-based event dispatching system may be ex-tended to cope with end user mobility and with the changing computational resources and network conditions of mobile systems.Our approach is to dynam-ically deploy and recon?gure the underlying event distribution infrastructure (i.e.,the‘broker’overlays).The approach is realized with the novel abstraction of opportunistic event channels.Their implementation involves dynamically cre-ated event brokers,the runtime installation of event modulators in brokers,and the dynamic relocation of such modulators.Adaptation policies are realized by adaptors that interact with a runtime resource monitoring system.Essentially, opportunistic channels implement a middleware-level analogue of the channel hando?protocol used in wireless communications.More importantly,by coor-dinating network-with middleware-level hando?s,opportunistic channels can attain substantial performance improvements over non-adaptive event channels. Such improvements are due to their use of shorter network paths and the better balancing of loads across event brokers.
Future work should address some de?ciencies of our current implementation, as well as generalize upon the basic concept of opportunistic channels.First, the current implementation assumes that there is only one intermediate broker between a provider and a consumer and therefore,cannot handle multi-level bro-ker topologies.We are extending our implementation and algorithm to address this limitation.Second,our current experimentation uses wireless networks that employ base stations,with only the last hop being a wireless link.Future work will use opportunistic channels over ad-hoc wireless networks.In addition,the current implementation assumes a reliable network environment and therefore, does not consider dynamic disconnection and reconnection.Our intent is to add application-speci?c failure recovery to brokers,an example being the mirroring and failure recovery described for business transaction systems in[10]. References
1. D.G.Andersen,H.Balakrishnan,M.F.Kaashoek,and R.Morris.Resilient over-
lay networks.In Proceedings of the18th ACM Symposium on Operating Systems Principles(SOSP-18),October2001.
2. F.E.Bustamante.The Active Streams Approach to Adaptive Distributed Applica-
tions a nd Services.PhD thesis,Georgia Institute of Technology,2001.
3. A.Carzaniga,D.S.Rosenblum,and A.L.Wolf.Achieving scalability and expres-
siveness in an internet-scale event noti?cation service.In Proceedings of the Nine-teenth Annual ACM Symposium on Principles of Distributed Computing(PODC 2000),pages219–227,Portland,Oregon,July2000.
4.M.C.Chan and R.Ramjee.Tcp/ip performance over3g wireless links with rate
and delay variation.In Proceedings of The Eighth ACM International Conference on Mobile Computing and Networking(MobiCom2002),Atlanta,GA,September 2002.
5.G.Chen and D.Kotz.An open platform for context-aware mobile applications.In
Proceedings of the First International Conference on Pervasive Computing(Per-vasive2002),August2002.
6.Y.Chen,K.Schwan,and D.W.Rosen.Java mirrors:Building blocks for re-
mote interaction.In Proceedings of the2002International Parallel and Distributed Processing Symposium(IPDPS2002),April2002.
7.G.Cugola,E.D.Nitto,and A.Fuggetta.The”jedi”event-based infrastructure
and its application to the development of the opss wfms.In IEEE Transactions on Software Engineering in2001,2001.
8.G.Cugola,E.D.Nitto,and G.P.Picco.Content-based dispatching in a mobile
environment.In In Workshop su Sistemi Distribuiti:Algorithms,Architecture e Linguaggi(WSDAAL),2000.
9.G.Eisenhauer,F.Bustamente,and K.Schwan.Event services for high performance
computing.In Proceedings of High Performance Distributed Computing-9(HPDC-
9),August2000.
10. A.Gavrilovska,K.Schwan,and V.Oleson.A practical approach for‘zero’down-
time in operational information systems.In Proceedings of the22nd International Conference on Distributed Computing Systems(ICDCS-22),July2002.
11.R.Gruber,B.Krishnamurthy,and E.Panagos.The architecture of the READY
event noti?cation service.In Proceedings of the19th IEEE International Con-ference on Distributed Computing Systems Middleware Workshop,Austin,Texas, USA,May1999.
12.W.Gu,G.Eisenhauer,and K.Schwan.Falcon:On-line monitoring and steering of
parallel programs.Concurrency:Practice and Experience,10(9):699–736,August 1998.
13. E.Gustafsson and et al.Mobile IPv4Regional Registration.draft-ietf-mobileip-
reg-tunnel-05,IETF,September2001.
14.Q.He,,and K.Schwan.Iq-rudp:Coordinating application adaptation with
network transport.In Proceedings of High Performance Distributed Computing-9(HPDC-9),July2002.
15. A.Hills.Wireless andrew.IEEE Spectrum,36(6),June1999.
16.Y.Huang and H.Garcia-Molina.Publish/subscribe in a mobile environment.In
2nd ACM International Workshop on Data Engineering for Wireless and Mobile Access(MobiDE’01),Santa Barbara,California,USA,2001.
17. D.Johnson and C.Perkins.Mobility Support in IPv6.Internet Draft,IETF,
draft-ietf-mobileip-ipv6-12.txt(work in progress),September2000.
18. D.B.Johnson and D.A.Maltz.Protocols for adaptive wireless and mobile net-
working.IEEE Personal Communications,3(1):34–42,1995.
19. D.B.Johnson and C.Perkins.Route Optimization in Mobile IP.In Internet
Draft(work in progress),1998.
20.R.Katz and E.Brewer.The case for wireless overlay networks.In SPIE Multimedia
and Networking Conference,January1996.
21. A.Medina,https://www.doczj.com/doc/4312454092.html,khina,I.Matta,and J.Byers.Brite:An approach to universal
topology generation.In Proceedings of the International Workshop on Modeling, Analysis and Simulation of Computer and Telecommunications Systems-MAS-COTS’01,Cincinnati,Ohio,August2001.
https://www.doczj.com/doc/4312454092.html,100.The Net100Project-Development of Network-Aware Operating Systems.
https://www.doczj.com/doc/4312454092.html,.
23. B. D.Noble,M.Satyanarayanan, D.Narayannan,J.Tilton,J.Flinn,and
K.Walker.Agile application-aware adaptation for mobility.In Proceedings of the16th ACM Symposium on Operating Systems Principles(SOSP-16),October 1997.
24.L.Opyrchal,M.Astley,J.Auerbach,G.Banavar,R.Strom,,and D.Sturman.Ex-
ploiting ip multicast in content-based publish-subscribe systems.In Proceedings of IFIP/ACM International Conference on Distributed Systems Platforms and Open Distributed Processing(Middleware2000),April2000.
25. C.Perkins.IP Mobility Support.IETF,Request for Comments2002,Oct.,1996.
26.T.Phan,L.Huang,and C.Dulan.Integrating mobile wireless devices into the
computational grid.In Proceedings of The Eighth ACM International Conference on Mobile Computing and Networking(MobiCom2002),Atlanta,GA,September 2002.
27.M.Pierce,C.Youn,and G.Fox.The gateway computational portal:Developing
web services for high performance computing.In Proceedings of2002International Conference on Computational Science(ICCS2002),April2002.
28. C.Poellabauer and K.Schwan.Power-aware video decoding using real-time event
handlers.In Proceedings of the5th International Workshop on Wireless Mobile Mult imedia(WoWMoM),September2002.
29. C.Poellabauer,K.Schwan,and R.West.Coordinated cpu and event scheduling
for distributed multimedia applications.In Proceedings of the9th ACM Multimedia Conference,October2001.
30.S.Rajagopalan and B.Badrinath.An adaptive location management strategy for
mobile ip.In Proceedings of the?rst annual international conference on Mobile computing and networking(MobiCom1995),December1995.
31. B.Segall and D.Arnold.Elvin has left the building:A publish/subscribe noti?ca-
tion service with quenching.In Proceedings of A UUG97,September1997.
32. E.C.Shek,S.K.Dao,Y.Zhang,and etc.Intelligent information dissemination
services in hybrid satellite-wireless networks.ACM Mobile Networks and Applica-tions(MONET)Journal,5(4),December2000.
33.P.Sutton,R.Arkins,and B.Segall.Supporting disconnectedness-transparent
information delivery for mobile and invisible computing.In CCGrid2001IEEE International Symposium on Cluster Computing and the Grid,May2001.
34.R.West,K.Schwan,and C.Poellabauer.Scalable scheduling support for loss
and delay constrained media streams.In Proceedings of5th IEEE Real-Time and Embedded Technology and Applications Symposium(RTAS1999),June1999. 35.H.Yokota,A.Idoue,and T.Kat.Link layer assisted mobile ip fast hando?
method over wireless lan networks.In Proceedings of The Eighth ACM Interna-tional Conference on Mobile Computing and Networking(MobiCom2002),Atlanta, GA,September2002.
36.X.Zhao,C.Castelluccia,and M.Baker.Flexible network support for mobility.In
Proceedings of The Fourth ACM International Conference on Mobile Computing and Networking(MobiCom1998),October1998.
37. D.Zhou,S.Pande,and K.Schwan.Method partitioning-runtime customization
of pervasive programs without design-time application knowledge.In Proceedings of the23nd International Conference on Distributed Compu ting Systems(ICDCS-
23),2003.
38. D.Zhou,K.Schwan,G.Eisenhauer,and Y.Chen.Supporting distributed high
performance application with java event channels.In Proceedings of the2001 International Parallel and Distributed Processing Symposium(IPDPS2001),April 2001.
82 Désignation Designation Bezeichnung Dimensions Abmessungen Dimensions de construction Dimensions for detailing Konstruktionsma?e Surface Oberfl?che G kg/m h mm b mm t w mm t f mm r 1mm r 2mm A mm 2d mm ? e min mm e max mm A L m 2/m A G m 2/t y s e s s 45o r 1 b r 2Fers U normaux européens Dimensions: DIN 1026-1: 2000, NF A 45-202 (1983)Tolérances: EN 10279: 2000 Etat de surface conforme à EN 10163-3:2004, classe C, sous-classe 1 European standard channels Dimensions: DIN 1026-1: 2000, NF A 45-202 (1983)Tolerances: EN 10279: 2000 Surface condition according to EN 10163-3:2004, class C, subclass 1 *Tonnage minimum et conditions de livraison nécessitent un accord préalable.* Minimum tonnage and delivery conditions upon agreement. * Die Mindestmengen pro Bestellung sowie die Lieferbedingungen sind im Voraus zu vereinbaren. 2006-1-3 h ≤300h >300b 2 b - t w 2 8%Inclinaison des ailes Flange slope Flanschneigung u 5%
外文翻译 原文 Material Source: Marketing Channels Author: Bert Rosenbloom Most producers do not sell their goods directly to the final users.Between them stands a set of intermediaries performing a variety of functions.These intermediaries constitute a marketing channel 1.The scope of Marketing channels: Marketing channels are sets of interdependent organizations involved in the process of making a product or service available for use or consumption..(MarketingKotler P., Asplund C., Rein I. and Haider H.D. 1999, Marketing management) 2.Marketing-channel decisions are among the most critical decisions facing management.The channels chosen intimately affect all the other marketing decisions.The company's pricing depends on whether it uses mass merchandisers or high-quality boutiques.The firm's sales force and adertising decisions depend on how much training and motivation dealers need.In addition,the company's channel decisions involve relatively long-term commitments to other firms.(MarketingKotler P., Asplund C., Rein I. and Haider H.D. 1999, Marketing management) 3.Empirical study in the channel for an industrial product found the effect of channel member satisfaction over time to be mediated by the extent of manifest conflict. The impact of manifest conflict was found to be mediated by the degree of channel member satisfaction. In other words, channel member satisfaction and manifest conflict within the channel were both antecedents and consequences of each other.( James R. Brown, Robert F. Lusch, Laurie P. Smith. Conflict and Satisfaction in an Industrial Channel of Distribution.1991.) 4.Intermediaries.Intermediaries normally achieve supperior efficiency in makng goods widely available and accessible to target markets.Through their contacts,experience,specialization,and scale of operation,intermediaries usually offer the firm more than it can achieve on its own.According to Stern and El-Ansary:Intermediaries smooth the flow of goods and services .This procedure is necessary in order to bridge the discrepancy between the assortment of goods and
Unit 9 Channels of Distribution Quiz Class _________ Name ________ Score __________ I. Translate the following terms. 1. 实体分销 6. 存货控制 2. 特许连锁7. 双轨销售 3. 配送中心8. 销售代理商 4. 专业店9. 仓储商店 5. 佣金行10. 折扣店 II．Decide whether the following statements are true or false. 1.With a relatively small number of potential customers, a manufacturer may use its own sales force to sell directly to consumers or industrial users. 2.When products of low unit value are sold in large quantities or are combined with other goods so that the total order is large, shorter channels may be economically feasible. 3.An industrial product that is highly technical is often distributed indirectly to industrial users. 4. A business with adequate finances can establish its own sales, grant credit, or warehouse its own products. A financially weak firm has to use middlemen who could provide these services. 5. A manufacturer is likely to use multiple channels to reach different markets when selling. 6.Logistics is totally different from physical distribution. 7.An independent retailer operates only one outlet and offers personal services, a convenient location and close customer contact. 8.Wholesaling is concerned with the activities of those persons or establishment which sell to retailers and other merchants, and industrial or commercial users, but do not sell in large amounts to final consumers. 9.Agent middlemen never actually own the goods, but they do actively assist in the transfer of title. 10.Selling agents take over the whole marketing job of producers, not just the selling function. III．Choose the best answer. 1.Which of the following is not a retailer? A.a supermarket B. a department store C. a discount store D.a rack jobber 2.________ provide a place where buyers and sellers can come together and complete a transaction.
CONNECTION SYSTEMS JORDAHL ? Anchor Channels JTA ETA-09/0338 Anchor Channels SHEAR REINFORCEMENT SYSTEMS SHEAR CONNECTOR SYSTEMS FA?ADE CONNECTION SYSTEMS MOUNTING TECHNOLOGY
Diese Zulassung umfasst This Approval contains 37 Seiten einschlie?lich 27 Anh?nge 37 pages including 27 annexes Diese Zulassung ersetzt This Approval replaces ETA-09/0338 mit Geltungsdauer vom 28.02.2012 bis 15.02.2015 ETA-09/0338 with validity from 28.02.2012 to 15.02.2015 E u r o p ?i s c h e O r g a n i s a t i o n f ür T e c h n i s c h e Z u l a s s u n g e n English translation prepared by DIBt - Original version in German language Handelsbezeichnung Trade name Jordahl-Ankerschiene JTA Jordahl-anchor channel JTA Zulassungsinhaber Holder of approval JORDAHL GmbH Nobelstra?e 51 12057 Berlin DEUTSCHLAND Zulassungsgegenstand und Verwendungszweck Ankerschienen Generic type and use of construction product Anchor channels Geltungsdauer: Validity: vom from 17 June 2013 bis to 17 June 2018 Herstellwerk Manufacturing plant 14959 Trebbin, Industriestr. 5 European Technical Approval ETA-09/0338 (O H F W U R Q L F F R S \ R I W K H (7$ E \ ',%W (7$
1.1 在eNodeB下进行实时性能监控和测试 在“信令跟踪管理”界面下，还可以进行eNodeB传输性能、小区性能、用户性能和RRU性能的监控和测试。 图 1 eNodeB性能监控和测试功能 1.1.1监控小区性能 小区性能监控功能主要监控项有业务满意率监控、总吞吐量监控、业务数/用户数监控、RB使用情况监控、RSSI统计监控、ICIC监控、虚拟MIMO监控、干扰检测监控等，DBR统计和被调度用户统计。 小区性能监控任务登记，需要设置被监控小区的Local Cell ID，可以设置监控周期和文件保存的路径，文件保存的格式有“csv”和“mmf”两种。 图 2 小区性能监控
常用小区性能监控项有总吞吐量监控、用户数监控、RB使用情况监控和RSSI统计监控。 1.1.2监控扇区性能 扇区监控主要监控上行宽频扫描功能 图 1 扇区监控性能 1.1.3传输监控性能 传输性能监控主要包括IP链路监控、IP PATH性能监控、IP性能监控、SCTP性能监控、UDP 灌包测试监控、本地流过路流监控和资源组监控。 图 2 传输监控性能 1.1.4用户监控性能 用户性能测试功能主要监控项有下行RSRP/RSRQ监控、误码率监控、Power Headroom监控、信道质量监控、调度监控、RLC业务量监控、吞吐量监控、AQM监控、上行功控监控、下行功
控监控、上行ICIC监控和按MCS阶数统计监控。 图 3 用户性能测试 用户性能测试功能任务登记，需要选择被测量的基站（站点数目最多可以选择30个），设置跟踪用户的信息，监控周期和文件保存的路径，文件保存的格式有“csv”和“mmf”两种。 1.1.5R RU监控性能 该任务用于监测RRU输出功率和温度的性能状况，每个RRU最多能启动的监测任务为：●一个输出功率监测任务（注：SPC310之前的版本该项监控不准） ●一个温度监测任务 图 4 RRU监控性能
营销渠道Marketing Channels 营销渠道Marketing Channels是指促使产品或服务顺利地被使用或消费的一整套相互依存的组织。也就是说，一条市场销售渠道包括某种产品的供产销过程中所有的企业和个人，如资源供应商、生产者、商人中间商、代理中间商、辅助商以及最后消费者或用户等。 其特点为： ①销售渠道主要是由参与商品流通过程的各种类型的机构组成的。通过这种机构组成的网络，产品才能进入市场流通销售，从生产者流向最终消费者或用户，实现商品的价值。 ②销售渠道的起点是生产者，终点是通过生产消费和个人生活消费能实质上改变商品形态、使用价值和价值的最后消费者和用户。 ③在商品从生产者流向最终消费者或用户的流通过程中，最少要经过一次商品所有权的转移。 ④销售渠道并不是生产者和中间商之间相互联系的简单结合，而是企业之间为达到各自或共同目标而进行交易的复杂行为体系和过程。 销售渠道的成员包括制造商、制造商设立的销售机构、代理商、分销商、经销商、批发商、零售商、消费者、用户、运输商、广告商、咨询商和服务商等。 渠道商概念要广于分销商。 制造商Manifacturer是生产制造商品的组织机构。 代理商Agency是指某产品在销售过程中由生产厂家授权在某一区域有资格销售该产品的商业单位。可分为全球级、国家级、地区级、省市县级等，又分为独家代理、总代理、分级代理等，所有代理商家都有相应的特权，低级别代理商原则上是由高一级的代理商管理。代理是代企业打理生意，不是买断企业的产品，不是自己用产品，是代企业转手卖出去，是厂家给额度的一种经营行为，货物的所有权属于厂家，而不是代理商家。所以代理商一般是指那些赚取企业代理佣金的商业单位。 代理商的建立，可以分担厂商的风险，使厂商与代理商共同拉动市场从而降低厂商的经营风险。 在代理商的层次上，除设立总代外，代理商还可以根据厂商的渠道模式，下设一级代理 或区域代理并同时与终端销售商合作。这样，代理商从简单的分销转换成具有管理职能 的渠道维护者，除业务管理外，代理商同时具备品牌管理、促销管理、服务对接、财务 管理等各项职能。 代理制和经销制也能互相渗透，各类销售渠道的主要区别在于代理商或经销商层级的增多。目前，厂商对于代理商和经销商的管理主要侧重于价格和货源的管理，对于人员培 训及广告等还会提供一些支持。对于代理商和经销商而言，其利润则主要来自于销售利润、返利、安装和维修服务等。 分销商Distributor广义理解是指那些专门从事将商品从生产者转移到消费者的活动的机构和人员，当这些分销商的活动产业化以后，分销业也就形成了。涵盖代理商和经销商。狭义理解是介于代理商和经销商之间的商业单位。随着批发概念的落伍，时髦的概念是分销，所谓的分销是分着来销。可见在商品销售的过程中，已经考虑到了下家的情况，不是盲目销售，而是有计划有策略地销售，商家有服务终端的概念。分销和批发是相对的，是从管理和计划
AC Alternating Current 交流电 ACB Alarm Collection Board 告警收集板 ACCH Associated Control Channel 相关控制信道 ACCU Alternating Current Connection Unit 交流电接线单元 A/Dconverter Analog to Digital converter 模数转换器 AGCH Access Grant Channel 接入准许信道 Air conditioner (Active cooler) One version of the climate unit.（有源冷却器）空调设备 AIS Alarm Indication Signal 告警指示信号 ALNA Antenna Low Noise Amplifier 天线低噪放大器 AM Application Module 应用模块 AMPS Advanced Mobile Phonesystem 高级移动电话系统 Antenna Diversity 天线分极接收 Adaptive Equalizatio 适应性均衡 AO Application Object 应用对象 Aoc Advice Of Charge 计费通知 APSI Application Platform Services Interface 应用平台业务接口 ARAE Antenna Related Auxiliary Equipment 天线相关辅助设备 Architecture 体系结构 ARFCN Absolute Radio Frequency Channel Number 绝对无线电频率信道数 ARU Active Replaceable Unit 有源可替换单元 ASIC Application Specific Integrated Circuit 特定用途集成电路 Astra ASIC in the TRU TRU中的一种特定用途集成电路