20 I 0 3rd International Conference on Advanced Computer Theory and Engineering(ICA CTE)

The Application of the Structure Designing Distributed Algorithm on the Mobile Computing Networks

HUANG Tian-kai, HUANG Sheng-Zhong

Department of Mathematics and Computer Science, Liuzhou Teachers College, Liuzhou, Guangxi 545004, China gxhsz@126.com, htk3000@163.com

Abstract-In order to acquire continuous network connectivity for client without considering the physical location of the user, a new structure designing distribution algorithm on the mobile computing network was established. The design distribution algorithm could account for the host mobility and physical constraints. Firstly, the development of mobile network was introduced. Then the system model was established. And the structure designing distributed algorithms were put forward. Then the search strategy and inform strategy were constructed respectively. The application of this method in actual mobile network was summary at last.

Keywords-mobile computing network; designing distribution algorithm; application

I. INTRODUCTION

Wide spectrums of portable, personalized computing devices, ranging from laptop computers to handheld personal digital assistants, had recently been introduced. Their explosive growth had sparked considerable interest in providing continuous network coverage to such mobile hosts (MH), regardless of their location. Mobile hosts had primarily been used as 'virtual desktops' enabling remote access to information stored at fixed hosts, e.g. electronic mail and messaging services. However, examples of collaborative applications between mobile clients had begun to form as well, and manipulating the state of an electrical network by field engineers equipped with handheld mobile machines. It had been predicted that with the proliferation of personal, mobile computers, advanced techniques would be need to control shared information distributed in such computers.

The design of distributed algorithms and protocols had conventionally been depended on an underlying network structure making up of not mobile users, i.e. the location of a host within the network did not change. Thus, in the absence of site and link failures, the connectivity amongst hosts in the network remains unchanged. Distributed algorithms thus assume a model comprising a set of processes executing on static hosts that communicate by messages over point-to­point logical channels. Each channel might span multiple physical links of the network; the set of links and the hosts at the endpoints of the channel did not change when the time past. But, this model was unable to capture the characters and constraints of mobile users, and the distributed algorithms aimed at this model therefore needed to be founded again to deal with user mobility.

To facilitate continuous network coverage for mobile users, a static network was augmented with mobile support

stations (MSS) that were each able to communicate directly with MHs within a limited geographical area which also named as cell, usually through a low-bandwidth wireless medium. In effect, MSSs served as access points for a MH to joint with the static network, and the cell, from which a MH connected to the static network, represented its current 'location'. MHs were thereby able to connect to the static part of the network from every location at any times. Firstly, the whole network topology transferred dynamically as MHs moved from one cell to another. This implied that distributed algorithms for a mobile computing environment could not assume that a user hold a fixed and universally known location in the network at all times; a mobile user should first be located before a message could be delivered to it. Further, as users transferred their locations, the physical connectivity of the network changed. Therefore, any logical structure, which many distributed algorithms exploited, could not be statically mapped to a lot of physical connections within the network. Secondly, the bandwidth of the wireless link connecting a MH to a MSS was important lower than the (,wired') links between static users. Thirdly, mobile users had tight constraints on power consumption relative to desktop machines, since they usually operated on stand-alone sources such as battery cells. Consequently, they often operate in a 'doze mode' or voluntarily disconnect from the network. Lastly, transmission and reception of messages over the wireless link also consumes power at a MH, and so distributed algorithms need to minimize communication over the wireless links. These aspects are characteristic of mobile computing, and needed to be considered in the design of distributed algorithms.

II. THE SYSTEM MODEL

The term 'mobile' implied that was able to move while retaining its network connections. A user that could move while retaining its network connections was a mobile host (MH). The infrastructure devices that communicate directly with the mobile hosts were called mobile support stations (MSS). A 'cell' was a logical or geographical coverage area under a MSS. All MHs that had recognized themselves with a particular MSS were considered to be local to the MSS. A MH could directly communicate with a MSS (and vice versa) only if the MH was physically located within the cell services by the MSS. At any given instant of time, a MH might (logically) belong to only one cell; its current cell defines a MH's 'location'. In this research, we assumed that all hosts and communication links were reliable. Further, for simplicity of presentation, we assumed that all fixed hosts

978-1-4244-6542-2/$26.00© 2010 IEEE V6-155

2010 3rd International Conference on Advanced Computer Theory and Engineering(ICACTE)

act as MSSs, and applied the terms MSS and 'fixed host' interchangeably.

The system model was shown in Figure 1, which was made up of two distinct sets of entities: many mobile hosts and relatively fewer, however more powerful, fixed hosts (MSS). The number of MSSs would be denoted by Nmss and that of MHs by Nmh with Nmh» Nmss All fixed hosts and the communication paths between them constructed the static/fixed network. A MSS communicated with the MHs within its cell through a wireless medium. The overall network architecture thus was made up of a 'wired' network of fixed hosts that connected the otherwise isolated, low­bandwidth wireless networks, each consisting of a MSS and the MHs local to its cell. Host mobility was represented in this model as migration of MHs between cells.

";rdess cell ";reloss cell

WIRED NETWORK FIXED HOST

MSS

";reless cell ";rdess cdl

Figure 1 The diagram of system model The static network offered reliable, sequenced delivery

of information between any two MSS, with arbitrary message latency. At the same time, the wireless network within a cell ensures FIFO delivery of information between a MSS and local MH. If a MH did not keep away its cell, then information sent from the local MSS would be received in sequence by the MH. But, since a MH might leave its cell at any time, the sequence of messages acquired at the MH was a prefix of the sequence of messages sent from the MSS. Therefore, a MH was needed to send leave information on the MH to MSS path offering the sequence number r of the last message got on the MSS-to-MH path. After sending this message, the MH neither sends nor receives any other message within the current cell. Each MSS kept up a list of ids of MHs that were local to its cell; on receipt of leave 0 from a local MH, it is deleted from the list. Conversely, when a MH entered a new cell, it sent a join (mh-id) to the new MSS; it was then added to the list of local MHs at the new MSS.

Information communication from a MH hI to another MH h2 happened as follows. The first sent the information to its local MSS MI by the wireless link. MI forwarded it to M2, the local MSS of h2 through the fixed network. M2 then transferred it to h2 through its local wireless network. But, because the location of a MH varied with every move and its current location was not universally known in the network, M required to first deciding where h2 was located before It could forward the message from hI to MZ. This was an important problem faced by network-layer routing protocols for mobile hosts. The system model was not linked to any

particular routing program, and instead we assumed that any information destined for a mobile host incurred a fixed search cost.

A typical measure of efficiency of a distributed algorithm for fixed networks was the communication complexity of the algorithm, and the number of information exchanged in one execution of the algorithm. However, with the introduction of mobile hosts and their associated resource constraints, the communication complexity should also conclude the search cost, and the number of information exchanged within the fixed network to locate a mobile host. Further, since MHs had tight constraints on power consumption, and messages sent on the wireless links require MHs to expend power, the communication complexity of a distributed algorithm for MHs should explicitly include the number of wireless messages. Thus, our system model concluded three cost measures for counting the number of messages exchanged:

C fixed - cost of sending a point-to-point message

between any two fixed hosts.

CWireless - cost of sending a message from a MH to its

local MSS over the wireless channel (and vice versa).

. Csearch - cost incurred to locate a MH and forward a

message to its current local MSS, from a source MSS. A typical search strategy for a MSS would be to query all other MSSs within a predefined coverage area to determine if a MH was local to its cell. The MSS currently local to the MH will respond, and the original MSS could then forward a data packet to this MSS. Using such a search strategy, where the search cost was linearly proportional to the number of

locations (MSS), Csearch is equal to N mss xC fixed. Based on the above cost parameters, information sent

from an MH to another MH incurred a

cost CWireless X Csearch' while information sent from a MSS

to a non-local MH incurred a cost CWireless + C.\'earch.

III. STRUCTURING DISTRIBUTED ALGORITHMS

We cast distributed systems with mobile hosts into a two-tier structure: (1) a network of fixed hosts with more resources in terms of storage, computing and communication; and (2) mobile hosts, which might operate in a disconnected or dozed mode, connected by a low-bandwidth wireless connection to this network. The guiding principle for structuring distributed algorithms for MHs in this model was that the computation and communication requirements of an algorithm should be suit for the static part of the system to the extent possible. Below, we offered justifications for this choice.

A message sent from a MSS to a non-local mobile host incurred a search cost. The same was also real for a message exchanged between two mobile hosts in different cells. To decrease the search component of the whole execution cost, it was desirable that communication between a fixed host and a mobile host happened locally within the same cell.

The ability of a MH to operate while on the move needed a stand-alone source of power, viz. batteries. Given the

V6-156

2010 3rd International Conference on Advanced Computer Theory and Engineering(ICACTE)

limited life of batteries, power consumption was a serious practical consideration at a MH. In addition to disk accesses and CPU operations, a MH had to expend its limited power resources to send and receive wireless information; such a constraint was not faced by information exchanged between fixed hosts over the wired network. Additionally, wireless paths had an important lower bandwidth than those within the fixed network. Thus, the number of wireless information exchanged in any algorithm execution should be minimal, possibly at the expense of a higher number of messages exchanged within the fixed network.

The two points mentioned above suggest that the data structures encapsulating the 'state' of an algorithm execution should reside at the fixed hosts; thus, information formed to update these data structures would be addressed to fixed hosts. Communication necessary to execute an algorithm might be split into three components:

The global component was made up of information whose source and destination was both fixed hosts. to update appropriate data structures, and mostly represented the communication necessary for the progress of an algorithm execution. The local component referred to communication within a single wireless cell between a MH and its local MSS, and would be used to initiate an algorithm execution from a MH or to communicate the final outcome of an execution from a MSS to a local MH. The search component was made up of information that the fixed hosts exchanged to decide the present location of a MH so that information addressed to this MH might forward to the appropriate MSS. Thus, our approach advised that the global part occupy the whole communication.

The two unique modes of operation of mobile hosts viz. disconnected and 'doze-mode' offered compelling arguments against executing an algorithm directly on MHs. When operating in a doze-mode, the MH shuts/slows down most of its system functioned to decrease power consumption, and only listens for incoming information. Like disconnection, this was a voluntary operation. However, the implications were different. In doze mode, a mobile host was reachable from the rest of the system, and thus could be induced by the system to resume its normal operating mode, if needed. By comparison, a disconnection and subsequent reconnection was initiated from the mobile host; it was cut off from the system in the intervening period.

A distributed algorithm designed for the mobile computing environment should not require each MH to participate in every execution of the algorithm. Otherwise, it offered those MHs from operating in a doze-mode that neither initiated the computation nor was the outcome of an execution significant to them, and consequently, wanted to conserve power by operating in doze-mode were completely thwarted. Thus, by downloading most of the communication and computation requirements to the fixed part of the network, the static hosts were responsible for the progress of an algorithm execution, and a mobile host would not be required to intervene unless it was interested in the outcome of the execution.

- Algorithms that directly executed at the mobile hosts need to consider the possibility that one or more of the

partIcIpants might disconnect while an execution of the algorithm is in progress. This had a two-fold effect:

(1) The algorithm should be designed to handle a variable number of participants while an execution was in progress;

(2) A search overhead would be incurred if the remaining (mobile) participants required to be informed of a MH's disconnection. Though distributed algorithms for static systems that were designed to be fault-tolerant did handle changes in the number of participants, it was inefficient to tackle disconnections of mobile hosts using these algorithms. Disconnection was a voluntary operation, and therefore a MH maight inform the system prior to an impending disconnection: thus, disconnection should not be associated with the same semantics as failure. The two-tier principle makes it easy to handle disconnections: since the fixed hosts are responsible for the progress to an algorithm execution, disconnection of one or more MHs did not alter the number of participants in the algorithm. Further, prior to disconnecting from the network, a MH could download any data to the fixed network that was necessary for progress of the algorithm.

IV. SEARCH STRATEGYE

(1) Actions executed by a MSS M On receipt of a request for the token from a local MH, M

added the request to the rear of its request queue. When M receives the token from its predecessor in the

logical ring, it executed the following steps: 1) Pending requests from M's request queue were moved

to the grant queue. 2) Repeat - Remove the request at the heat of the grant queue. - If the MH making the request was currently local to M,

then deliver the token to the MH over the wireless link. - Else, search and deliver the token to the MH in its

current cell. - Await return of the token from the MH. Until grant

queue was empty. 3) Forward token to M's successor in the logical ring. (2) Actions executed by a MH h 1) When h needed access to the token, it submitted a

request to its current local MSS. 2) The MSS where h submitted its request will

eventually sent the token to h. After h accesses the critical region, it returned the token to the same MSS.

The above algorithm assumes that a MH did not submit a second request if its previous request had not yet been serviced. Secondly, when a MH receives the token, it should return it to the sender MSS after accessing the critical region, i.e. it might not disconnect permanently after receiving the token. Correctness sketch Mutual exclusion was trivially guaranteed by the algorithm, since at most one MH may hold the token at any given time. Next, consider why starvation did not happen, i.e. every request submitted by a MH was eventually granted. It required to be shown the token could not reside forever at a fixed host, and thus it eventually visited every fixed host in the ring. First, mote that the maximum number of requested serviced by a MSS

V6-157

2010 3rd International Conference on Advanced Computer Theory and Engineering(ICACTE)

holding the token was bounded: only those requests that were made prior to arrival of the token (since the token's last visit) are serviced. When the token arrived at a MSS, contents of the request queue are transferred to the grant queue; new requests were then added to the request queue. Only those requests that belong to the grant queue were satisfied by the MSS. It followed that the grant queue

concluded at most N mh requests, one per MH.

V. INFORM STRATEGY

(I) Actions executed by a MSS M On receipt of a request from a local MH h, M added a

request < h, M > to the rear of its request queue.

Upon receipt of a inform < h, M' > message, the

current value of loen( h) is replaced with M'in the entry

< h,loen(h) > in M's request queue.

On receipt of the token, M executes the following steps: I) Entries from the request queue were moved to the

grant queue. 2) Repeat - Remove the request < h, loen( h) > at the

head of the grant queue

- If loen(h) M, then deliver the token to h over the

local wireless link - Else, forward the token to loen(h) , i.e. the MSS

currently local to h, which would transmit it to h over the local wireless cell.

- Await return of the token from h. Until grant queue was empty 3) Forward token to M's successor in the logical ring. (2) Actions executed by a MH h When h needed access to the token, 1) it submited a request to its current local MSS, say M,

and 2) stored M in the local variable req _loen(h) . When h receives the token from the MSS req _loen ,

it accessed the critical region, returned the token to the same

MSS and then sets req _loen to 1.

After every move, h now included req _loen with the

joinO message, i.e. it sent joint (h, req _loen ) message to

the MSS M' upon entering the cell under M' .

- If req _loen received with the joinO message was

not -L, then M' sends a inform (h,M') message to the

MSS req _loen . Comparison of search and inform strategies. To compare

the search and inform strategies, let an MH h submited a request at MSS M and received the token at M'. Assume that it made MOB number of moved in the intervening period. Then, after each of these moved, a inform 0 message was

sent to M, i.e. the inform cost was MOB X C fised' Note that

the inform 0 message might be combined the join 0 message when the MH entered a cell; thus, there was no

additional CWireless component to the overall inform cost. In

algorithm R-MSS: search, on the other hand, M would search for the current location of h and the cost incurred

would be Csearch' Thus, the inform strategy is preferable to

search strategy when MOB X C fised < C fixed' i.e. if h transferred cells 'less often' after submitting its request, then

it was better for h to inform M of every changed in its

location rather than M searching for h . The concept of designing distributed algorithms that

explicitly coped with host mobility was still in its infancy. An overview of the impact of host mobility on distributed computations was presented, while a scientist contained preliminary ideas that had been fully developed in this paper. The implications of mobile for distributed data management were considered elsewhere. Other related work included addressing schemes and protocols at the network-layer for routing messages to and from mobile hosts, while a specialist quantified the effects of host mobility on transport­layer connections. Application layer protocols to deliver multicast messages to mobile recipients from exactly-one location were presented. The design of distributed file systems had also been influenced by mobility of users. Various operating systems issued related to mobile hosts could be found elsewhere.

VI. CONCLUSIONS

The design of algorithms for distributed systems and their communication costs had been depended on the assumptions that the location of hosts in the network did not change, and the connectivity amongst the hosts was static in the absence of failures. However, with the emergence of mobile computing, these assumptions were no longer valid. Additionally, mobile hosts had tight constraints on power consumption, and bandwidth of the wireless links connecting MHs to their local MSSs was limited. This paper first presents a new system model for the mobile computing environment, and then described a general principle for structuring distributed algorithms in this model.

REFERENCES

[I] w.e. Grant, Wireless Coyote, "A computer· supported field trip," Comm. ACM, 1993(5).

[2] Y-K. Kwok, I. Ahmad, "Efficient scheduling of arbitrary task graphic to miltprocessors using a parallel genetic algorithm," 1. Parallel Distrib. 1997,47(1).

[3] W Boyer, G. Hura, "A new min-span heuristic algorithm for task scheduling in heterogeneous systems," Proceedings of the Sixth Biennial World Conference On Integrated Design And Process Technology. 2002.

[4] H. Topculoglu, S. Hariri, M.Y Wu, "Performance-effective and low complexity task scheduling for heterogeneous computing", IEEE Trans. Parallel Distrib. Systems, 2002, 13(3).

V6-158