A QoS-Aware Transcoding System Using Composite MultiMedia … · 2006. 9. 29. · A QoS-Aware Transcoding System Using Composite MultiMedia Document and Component Merge Queue Chung-Hsien

A QoS-Aware Transcoding System Using Composite MultiMedia Documentand Component Merge Queue

Chung-Hsien He and Ren-Song KoDepartment of Computer Science and Information Engineering

National Chung Cheng University, Chia-Yi, Taiwan 621

Abstract: - In recent years, most of the research work on transcoding proxies in mobile computing en-vironments mainly focus on designing the architectures of systems for achieving power conservation, highscalability, and high bandwidth utilization. However, fully transforming a document still causes extra burdenon the system and increases response time. We propose a document structure, composite multimedia doc-ument or CMD, and a new system architecture, component merge queuing transcoding proxy or CMQTP,to improve overall performance. Several simulations are conducted to measure the performance of CMQTPcooperating with CMD. The results show that the proposed architecture is more scalable than traditionalclient-server systems and is able to effectively control the system load to attain the desired QoS.

Key-Words: - QoS, Content Adaptation, Transcoding Proxy, Multimedia, Wireless Networks, Mobile Com-puting

1 Introduction

The advance in wireless communication enablesusers to access information systems anytime, any-where, via various mobile devices. Thus, con-tent providers are establishing a number of mo-bile services such as route guidance, location-dependent entertainment information, and weatherforecasting. However, the mobile devices, suchas laptops, personal digital assistants (PDAs) andsmart phones, differ significantly in their hardwarecharacteristics, software capabilities, and networkconnectivity. The high diversity in the capabili-ties of mobile devices brings many challenges tothe design of mobile information systems. Thesechallenges have been research issues in pervasivecomputing [6] including power conservation, highscalability, high bandwidth utilization and contextawareness [7].

Content adaptation [3], an important technique

to realize context awareness, emerges to remedy theproblem resulting from the said diversity by offer-ing the different mobile users suitable versions ofthe same object according to the capabilities of themobile devices or the network traffic. For instance,the information system will transmit the completeversion of the requested document to a high-end de-vice, but may omit all video clips and images to re-duce document processing requirements for a lesscapable device. However, there may be problemswith such a multimedia-omission approach. For in-stance, a document may describe geographical in-formation about a map, and the image of the map isvery important. Thus, the system should not omitthe image but reduce its resolution.

Furthermore, mobile users usually access net-works via wireless shared media. A popular doc-ument may be requested by many users simultane-ously; therefore, the bandwidth may be conservedby merging these requests and multicasting one

Proceedings of the 10th WSEAS International Conference on COMMUNICATIONS, Vouliagmeni, Athens, Greece, July 10-12, 2006 (pp92-97)

copy of the document to all users via shared me-dia. In this paper, we shall propose our system foraddressing the content adaptation and QoS issues.

2 Related Work

2.1 On-demand Data Broadcasting

In an on-demand data broadcasting system, aserver maintains a data request queue and servesthese requests according to the employed schedul-ing algorithm. A scheduling algorithm is used toprioritize all data requests in the data request queue,and the server will serve these data requests ac-cording to their priorities. Besides, requests for theidentical data object are merged and one copy ofthe data object is transmitted back for conservingbandwidth [1].

2.2 Content Adaptation System

Such a system decides the optimal content ver-sion for presentation and the best strategy for de-riving that version, and then generates that version.The most crucial component of the system is thedecision engine, which resides on the proxy serverfor negotiating the strategy [3].

2.3 Transcoding Proxy with a Queueing Net-work

Huang et al. [2] modeled a transcoding proxy asa queueing network containing three queues (con-trol channel, scheduler, and broadcast channel). Byanalyzing the queueing network, theoretical resultsmay be derived to formulate the average waitingtime of each queue. The average waiting time ofthe overall system will be equal to the summationof the average waiting time of the three queues.

3 Problem Formulation

In the past, a transcoding proxy, which trans-forms a data object from one version into another,is placed between a client and an information serverto coordinate the mismatch between what the server

provides and what the client prefers for realizingcontent adaptation [4][5]. Most of the researchwork on transcoding proxies in mobile computingenvironments mainly focus on designing the archi-tectures of systems for achieving power conserva-tion, high scalability, and high bandwidth utiliza-tion. Even though a new architecture may improveperformance successfully, fully transforming a doc-ument still causes extra burden on the system andincreases response time.

Furthermore, a popular document may be re-quested by many users simultaneously. It sim-ply wastes bandwidth to transmit back the identi-cal document to each user separately over wirelessshared media.

4 Problem Solution

We propose a scalable and QoS-aware transcod-ing proxy architecture by utilizing composite mul-timedia document structure, parallel componentmerge queues, and on-demand broadcasting tech-nique [1]. Explicitly, we first define a compos-ite multimedia document structure, abbreviated asCMD. Continuously, we design a new transcod-ing proxy architecture, component merge queuingtranscoding proxy or CMQTP, to manage user re-quests and components of multimedia documents.By on-demand broadcasting, requests for the iden-tical data object can be safely merged since onetransmission of the data object in a broadcast chan-nel is able to satisfy all requests.

By componentizing the document, mobile de-vices may select appropriate components based ontheir capabilities. More components will give moredetailed information about the document. There-fore, transcoding proxies need not transcode the en-tire document but only the required components forless system load. For instance, a document may beseparated into three components as shown in Fig. 1.The laptop may require three components for themost detailed description of the document, and thesmart phone may only require one component forthe brief description.

Furthermore, with CMQTP and on-demandbroadcasting, if the three devices in Fig. 1 request


Figure 1. CMD cooperates with CMQTPand on-demand data broadcasting

the identical document, the system will merge thetheir requests and only transmit one copy of eachcomponent instead of three Outline components,two Patch1 components, and one Patch2 com-ponent.

4.1 CMD Structure

There are several existing approaches, such asinformation summarization [8], and transcodingproxy, for the message services of information sys-tems over mobile devices. This section will de-scribe a document structure, CMD, which is com-posed of several components, usually one skeletonand several patches. The skeleton, usually but notlimited to text-based, is the brief of the document.Users can understand original intention of the doc-ument from the skeleton. Patches may be appliedto skeleton or other patches, and will give more de-tailed information about the content in various for-mats such as text, image and video.

When mobile users send requests to the informa-tion system for the identical multimedia document,the information system will first transmit one copyof the skeleton to the mobile devices with broad-casting technique so all users may browse the out-line of the document. Continuously, the informa-tion system will transcode other patches accordingto the capabilities of mobile devices and transfer

patches to capable devices to reconstruct the docu-ment from the skeleton and patches.

Note that CMD is just a concept and may be ap-plied to various multimedia document types and in-formation systems. Thus, it is possible to customizeCMD structure. In this paper, we propose a CMDstructure with XML scheme for the web server. Aweb page is usually created in HTML or DHTMLformat, so we extend the HTML format with addi-tional XML tags and insert them into web pages forimplementation of CMD.

A skeleton may be composed of sections, andauthors may use <section> tags to describe thebrief of the document with text only. They mayuse <patch> tags to include extra informationsuch as annotations, images, and video clips. The<patch> tag has an attribute type to indicatethe information format such as text, image, andvideo. The attribute name is used to indicate wherethe patch may be applied to. For example, in Fig. 2,the tag <patch> in the fourth line of Skeletonindicate the position where Patch1 will be ap-plied to since the values of both attribute namematch. Patches may be prioritized via the attributepriority to indicate their relative importance. Ifthe patch content is not worth bandwidth and com-putation, its priority should be lower. Besides, eachinformation format may have extra attributes to in-dicate the presentation quality. For instance, thetext format has the attribute count for the numberof character, and the image format has the attributesresolution and color.

Fig. 2 illustrates a CMD example of a web page.Fig. 3 shows how the document may be presentedon three kinds of mobile devices, namely, lap-top, PDA, and smart phone. Here the laptop willreceive all patches. The PDA is constrained bylack of memory so it omits Patch1 and Patch4.The smart phone is constrained by screen size ornumber of characters, so it ignores Patch2 andPatch5 additionally.

4.2 System Architecture

To further accord with the characteristic ofthe CMD structure, we propose a new transcod-


<html>

<body>

<section 1>

<patch name = "331-01">

</patch>

Internet Transcoding for Universal

Access

More and more pervasive devices are

gaining access to the Internet and other …

<patch name = "331-02">

</patch>

<patch name = "331-03">

</patch>

</section 1>

<section 2>

Introduction

In order to improve universal access, we

are developing a system for transcoding

multimedia and Internet content. The

system uses an InfoPyramid for managing

and

manipulating multimedia content composed

of video, images, audio and text. …

<patch name = "331-04">

</patch>

The InfoPyramid manages the different

versions of media objects with different ...

<patch name = "331-05">

</patch>

</section 2>

</body>

</html>

<patch name="331-01" priority=5 type=image

resolution=525*352 color=256>

<img height=352 width=525 src="./

uaone.gif">

<patch name="331-02" priority=3 type=text

count=876>

Enabling universal access of multimedia

content has become increasingly important...

To enable universal access in the coming age

of pervasive computing, we are developing …



<img height=54 width=600 src="./

masthead.gif">



<img height=352 width=525 src="./iptr-

1.gif">

<patch name="331-05" priority=4 type=text

count=271>

<LI>

A representation scheme called the

InfoPyramid that provides a multi-modal…

</LI>

<LI>

A customizer that selects the best content

representation to …

</LI>

Skeleton Patches

Patch1

Patch2

Patch3

Patch4

Patch5

Figure 2. CMD web page example

(a) Laptop (b)PDA (c)Smart Phone

Figure 3. CMD with different devices

Figure 4. CMQTP architecture

ing proxy architecture, component merge queuetranscoding proxy or CMQTP. Some research ontranscoding proxies use request queue to improveperformance of the system [2]. In addition to re-quest queue, our system uses component mergequeue to further reduce the bandwidth and responsetime.

Fig. 4 illustrates the structure of CMQTP. Here(iD, capability) denotes the ith device requestingfor information D with capability which may becategorized as laptop, PDA, smart phone, etc. Inpractice, techniques for automatically discoveringthe client device type, networking characteristics,and client identification might be used to generatethe necessary context information [3]. In this pa-per, we simply assume clients can piggyback devicetype information within user requests so the systemwill know the necessary context information in ad-vance.

After receiving user requests, CMQTP will an-alyze the request queue and merge requests of theidentical document with the same level of capabil-ity. Continuously, CMQTP will determine docu-ment components based on capability. Before be-ing sent out, the components will enter the compo-nent merge queue and wait for extra time. Withinthis time, CMQTP will have a chance to analyzethe component merge queue and merge the identi-


Determine components and

transcode components according

to devices’capabilities

(1D, P) (2D, L) (3D, P)

request merge

(4D, S)

L : Laptop

P : PDA

S : Smart Phone

C1~C7 : component set

C1 C2 C5 C2 C3C4 C3 C6C7

component mergecomponent merge

request queue

component queue

If additional waiting time is no up, wait

for some time to promote merge rate

If additional waiting time is

up, broadcast components

Figure 5. Process of request and compo-nent merge queue

cal components.Fig. 5 illustrates this process. Devices 1, 2, 3,

and 4 request for D with capability P, L, P, and Srespectively. CMQTP will merge the requests (1D,P) and (3D, P) because of the same capability P.After processing these requests, the system gener-ates document components. Instead of being di-rectly sent out, these components will stay in com-ponent merge queue for extra time so CMQTP haschance to further merge the same component gen-erated later within the extra time. In this example,the two same components C2 are merged, and soare the components C3.

5 Performance Evaluation

We developed a simulator to evaluate the perfor-mance of CMQTP with CMD. There is a multime-dia document composed of one skeleton and fivepatches as showed in Fig. 2 in the web server. Thesizes of Skeleton, Patch1, Patch2, Patch3,Patch4, and Patch5 are 2k-Byte, 12k-Byte, 1k-Byte, 4k-Byte, 7k-Byte, and 1k-Byte respectively.There are three different types of mobile devices,namely, laptop, PDA, and smart phone.

The additional waiting time of component merge

avg response time/per-request (no waiting time)

0

2000

4000

6000

8000

10000

12000

20 40 60 80 100 120 140 160 180 200

clients

min

i-se

cond

normal

merge

Figure 6. Average response time vs clients

queue of CMQTP is preset as 1 second. There areone hundred clients and each client may send a re-quest randomly during a five-second time span.

Fig. 6 shows the impact on average responsetime per request. Note that the normal transcodingproxy transforms a document from one version intoanother; thus, the response time increases severelywith the number of clients and requests. On theother hand, the CMQTP system merges identicalrequests, so the response time does not increase sig-nificantly.

Fig. 7 shows the impact on average bandwidthper request. Since the CMQTP system mergesidentical requests, the system may transfer lesscomponents and serve more mobile devices. Con-sequently, component merging with broadcastingtechnique can reduce average bandwidth.

Fig. 8 shows what impact the popularity of doc-uments may have on the system performance. Herethe popularity of a document is defined as the num-ber of requests for the document divided by thenumber of total requests. There are two differ-ent CMD documents which have similar numberof components and component size in the CMQTPsystem. When all users only request for one identi-cal document, the performance will be the greatestsince all requests are identical and the possibility ofthe components being merged increases. As usersrequest for two documents fairly, i.e., the highestpopularity of the documents decrease, the possibil-ity of components being merged in the component


avg bandwidth/per-request (no waiting time)

0

2000

4000

6000

8000

10000

12000

14000

16000

18000

20000

20 40 60 80 100 120 140 160 180 200

clients

byte

s normal

merge

Figure 7. Average bandwidth vs clients

merge rate vs performance (200 clients within 15 seconds)

0

1000

2000

3000

4000

5000

6000

7000

8000

100 90 80 70 60 50

the highest popularity of all documents

min

i-se

cond

&byte

s

response

bandwidth

Figure 8. The effect of documentpopularity

merge queue decreases, and, thus, response timeand bandwidth increases. In a CMQTP system,the worst case will happen when all documents areequally popular.

6 Conclusion

Traditional transcoding proxies transform fullya document from one version into another, andsuch an approach will bring extra burden to infor-mation systems and increase response time. To-gether with shared medium characteristics of wire-less networks, we propose a scalable and QoS-aware transcoding proxy architecture by utilizingcomposite multimedia document structure, parallelcomponent merge queues, and on-demand broad-casting technique to improve the performance oftransmitting multimedia documents over wireless

networks. Our simulation results show that the pro-posed architecture is more scalable than traditionalclient-server systems and is able to effectively con-trol the system load to attain the desired QoS.

Note that the extra waiting time in the compo-nent merge queue may have a significant impact onperformance. The longer the time is, the more pos-sibility CMQTP may have to merge components.With more components being merged, the responsetime may be reduced. However, the extra waitingtime delay the transmission of components, and di-rectly increase the response time. Thus, as our fu-ture work, we would like to investigate and modelthe relationship between extra waiting time and re-sponse time. We may design an adaptive approachso the extra waiting time may be automatically ad-justed based on popularity of the documents, cur-rent system load, and network traffic to achieveminimum response time.

References

[1] D. Aksoy and M. J. Franklin. Scheduling for Large-Scale On-Demand Data Broadcasting. In Proceed-ings of IEEE INFOCOM Conference, pages 651–659, March 1999.

[2] J.-L. Huang, M.-S. Chen, and H.-P. Hung. A QoS-Aware Transcoding Proxy Using On-demand DataBroadcasting. IEEE INFOCOM, 3(53):2050–2059,2004.

[3] W. Y. Lum and F. C. M. Lau. A Context-Aware De-cision Engine for Content Adaptation. IEEE Perva-sive Computing, 1(3), July-September 2002.

[4] R. Mohan, J. R. Smith, and C.-S. Li. Adapting Mul-timedia Internet Content for Universal Access. IBMT.J. Watson Research Center, 1999.

[5] R. Mohan, J. R. Smith, and C.-S. Li. TrancodingInternet Content for Heterogeneous Client Devices.IBM T.J. Watson Research Center, 1999.

[6] M. Satyanarayanan. Pervasive Computing: Visionand Challenges. IEEE Personal Communications,8(4):10–17, August 2001.

[7] P. Steggles, A. Ward, A. Harter, A. Hopper, andP. Webster. The Anatomy of a Context-Aware Ap-plication. In Proceedings of the 5th ACM/IEEE In-ternational Conference on Mobile Computing andNetworking, pages 59–68, August 1999.

[8] Y.-H. Tseng. A News Content Summarizer for Chi-nese Cell Phones. IEEE ICME, (53):85–88, July2005.


Documents

A QoS-Aware Transcoding System Using Composite MultiMedia … · 2006. 9. 29. · A QoS-Aware Transcoding System Using Composite MultiMedia Document and Component Merge Queue Chung-Hsien