You are on page 1of 12

Energy Consumption and Conservation in Mobile Peer-to-Peer Systems

BY: T.SRINIVASU III-MCA ROLL NO:10039

Because of ABSTRACT Today’s mobile devices are growing in number and computational resources. Each device is capable of storing Megabytes or Gigabytes of digital content. Computational and energy constraints. we perform a detailed study of energy consumption of a structured P2P overlay on a PDA device. today’s networked devices are taking us closer to the goal of ubiquitous computing . smart-phones. even MP3 players Have some type of wireless connectivity. these devices have been limited to data access and communication through centralized servers or proxies. Using we actual energy energy measurements. These wireless devices provide an attractive platform for deploying wireless peer-to-peer (P2P) applications such as streaming video over wireless. Virtually all mobile devices today. To copy otherwise. however. Recent Devices capable of storing gigabytes of digital content are becoming ubiquitous. Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are Not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. the always on communication patterns of P2P networks is not a natural fit for energyconstrained mobile devices. greatly limiting their mobility and application access to network resources. to post on servers or to redistribute to lists. Based on these observations. to republish. making them an ideal platform for peer-topeer content delivery and sharing.INTRODUCTION As they scale down in form factor and grow in network connectivity. content-sharing and storage. In this paper. we implement an approach to improve energy conservation on P2P protocols and show somepromising preliminary results. laptops. requires prior specific permission and/or a fee. However. present consumption results for different type of operations in P2P overlays. and real-time event notification. handheld PDAs. .

STRUCTURED P2P AND CHIMERA A structured peer-to-peer overlay is an application-level network connecting any number of nodes. First. called its root node. and our power model in Section 3. The paper makes three IP address or a secure public key. scalable multicast . P2P applications assume that peers are always on and available to route messages or satisfy queries for data. these devices are still severely constrained in energy. While . we study the feasibility of deploying structured overlays on mobile devices from the perspective of energy consumption. resources. To the best of our knowledge. we port a secondgeneration P2P protocol (Chimera) over to an embedded computing platform similar to The overlay dynamically maps each key to a unique live node. While hardware advances have led To improved computational and storage the HP iPAQ. Nodes are assigned nodeIds uniformly at random from a large identifier space. We describe our experimental methodology. These structured overlays have been shown to scale to millions of hosts while maintaining low per-node state and low maintenance traffic . we measure the energy overhead of P2P applications by comparing a P2P chat program against a client-server clone of MSN Messenger. each representing an instance of an overlay participant. we implement an approach to improve energy conservation for peer-to-peer protocols. Nodes can generate their nodeIds by applying a secure one way hashing function such as SHA-1 to either its In this paper. and scalable event notification . and present the results in Section 4. Peer-to-peer applications are likely to exacerbate this energy shortage. Applicationspecific objects are assigned unique identifiers called keys from the same space.advances in peer-to-peer overlays led to the development of Internet-scale distributed applications such as distributed storage . and present some promising preliminary results in Section 5. features and constraints of our target platform. Second. and quantify its energy consumption using a high-precision measurement platform. where the client can perform disconnected operations and utilize intermittent connectivity. Unlike existing clientserver applications. no work has studied the impact of this always-on communication model on energy-constrained mobile devices. contributions. Finally.

a key’s root can change with network membership. Routing proceeds by forwarding the message incrementally closer in the namespace to the desired key. Existing protocols share several important Our target board. To support a network of size N. A node’s leafset is a small number of nodes closest to it in the namespace. each node forwards the message using a locally maintained routing table of overlay links. prefix-based structured overlay protocol implemented in C as an application library. most protocols require per-node routing state that scales as O(logN) and provide worst case O(log N) overlay hops between any two nodes. and . using routing table links for “long hops” through the namespace. It builds routing tables using proximity routing techniques from Tapestry and uses per-node leafsets for stability across failures like Pastry . and is almost same as an HP iPAQ. at any given time in a consistent network. support a network of size N. using routing table links for “long hops” through the namespace. most protocols require per-node routing state that scales as O(logN) and provide worst case O(log N) overlay hops between any two nodes. Chimera is a light-weight Chimera is a light-weight prefix-based structured overlay protocol implemented in C as an application library. characteristics. To deliver a message based on a key to its root node (key-based routing [5]). Figure 1 shows an example of routing state for a single Chimera node (node 3001). It builds routing tables using proximity routing techniques from Tapestry and uses per-node leafsets for stability across failures like Pastry . The Star gate is a small sensor network Gateway running Linux. a single node is responsible for each key. A node’s leafset is a small number of nodes closest to it in the namespace EXPERIMENTAL METHODOLOGY Our goal is to perform detailed energy measurements of a handheld device while running a peer-to-peer protocol. Routing message proceeds by forwarding closer in the the incrementally namespace to the desired key.

4. The monitor start() function signals the start of profiling to the DAQ and records a timestamp. we use another A/D channel to sample the wireless card voltage. The Stargate runs Linux Kernel 2. These pins expose the power supply of Stargate and the wireless card. Setup and Trace Generation As our handheld platform. We built a user library to control access to GPIO port and record profile information. The profiled application calls monitor init to open the The Stargate provides debugging pins and input/output ports that are easy to interface. One challenge in our measurement design lies in correlating power data with the application runtime events on the Stargate. we use the Stargate embedded computer.correlate detailed fluctuations in energy with specific function calls executed on the device. To access the GPIO port. characterization was a significant challenge. We use this file in non-buffered mode to send set/clear commands to the GPIO port.19 as its operating system. which acts as an event id for the user to distinguish consecutive start/stop events in . we use its control file. we give a detailed description of our experimental setup and describe our measurement methodology on the Stargate embedded board. has no display. monitor stop function clears the line. The DAQ samples each channel concurrently at 1000 times per second. During power profiling.11b interface. and a 256MB Compact Flash card. we make a third connection between the DAQ device and a Stargate output interface (GPIO Port 26). 64 MB of RAM. Both start and stop functions take a character pointer as a parameter. a wireless Netgear 802. GPIO control file. The Stargate. resets the port and allocate a log buffer to store application power profile. In this section. a special file that is directly mapped to a kernel device driver. we use one A/D channel of the DAQ to sample the voltage across a resistor that is connected in series to the main power supply and the Stargate. which we use to monitor their energy consumption. shown in Figure 2. a popular lowpower sensor network gateway nearly identical in architecture to the HewlettPackard iPAQ. To address this. Our experiments use a Stargate device with a 400 MHz Intel XScale processor. Performing detailed energy Similarly. The library includes simple functions for controlling the monitor. /proc/platx/gpio/GPCTL. generating a power trace which we download for post analysis.

Furthermore. even when the system is totally idle. This device driver records all scheduling decisions into a ring that we allocate in kernel space. The scheduler traces that we capture provides this information. The threads in our overlay or application may be interrupted or blocked during system calls. acts as an interface to this buffer. To capture the scheduler trace without interfering with operating execution execution. To compute the power. A device file. Since we use a general-purpose operating system. On each scheduling decision. The application trace and the scheduler trace files both include timestamps generated by the Stargate device accurate to the level of microseconds. we combine the application. When the user reads this file. we record the We correlate the power trace and the application/kernel traces by using trigger events across the GPIO port to synchronize timestamp values between the Stargate and the DAQ. enough to capture all scheduler decisions for our monitoring period. After each experimental run. and the scheduling decision time in microseconds resolution. scheduler and power trace files to compute energy consumption per task and per function. we use this driver to redirect /proc/pmulog into a file for offline analysis. /proc/pmulog. As we described. continue to run at the background for regular maintenance. we first compute the . lightweight tasks. To accurately match up energy consumption with specific threads and functions. is timestamped using the DAQ clock at the granularity of milliseconds. many daemons. The context of the character pointer is transparent to the user library.the log trace. the device driver displays the latest ring contents in ASCII format. we developed a lightweight (less than %2 overhead) device driver. We call this data as application trace. we need to know when threads block and when context switches occur. We continuously log kernel scheduling decisions to capture energy behavior of operating system and application threads. Finally. The ring has 16K entries. The power trace however. These two traces are easily merged in time sequential order. monitor end resets the GPIO port and writes the application log buffer to a file. which are thread identifier. the DAQ samples the voltage on the resistor. Offline Analysis Once the data collection is over.

5). Node arrival is simulated using the Poisson process (λ = 0. First. Overlay Scenarios Our measurement includes three different application scenarios for the P2P overlay. We first discuss energy consumption rates of the wireless interface. 50. the mobile node actively sends and receives chat messages 1. Finally. we compare overall energy consumption running a decentralized chat client on Chimera against a traditional client-server implementation. since it is generally regarded as a significant consumer of energy in handheld devices. Each node is created with a predefined session time. scale 30). 1. after which the node leaves. According to specifi-cations. generated using the Pareto distribution (shape 0.25 watts in receive mode and 1. in the Control Overhead experiment. Previous research shows communication energy consumption is a linear function of the time the wireless card stays in transfer and receive states. Nodes are spread across the two laptops evenly. 100.e. Second.4 watts. we run all other Chimera overlay nodes on two laptop computers connected to our Stargate via wireless 802. Throughout these maximum power consumption in power savings mode is 0. and 200. we measure Chimera’s energy costs while joining and maintaining a structured overlay without any application traffic. our card’s operating scenarios to try to isolate different Chimera components. First. We take two orthogonal approaches to measuring energy consumption. Our node does not initiate any application activity and only forwards traffic for other peer who are exchanging chat messages.75 watts in transmit . We ran each scenario on networks of size 25.78. we present our measurement results in detail.11b connections in ad hoc mode. Each node randomly chooses another node to message after a time interval generated using the Pareto distribution (to model inter-message arrival time) . In the second Passive Participation scenario.current (i. we measure Chimera’s power consumption under different experiments. we try to isolate the energy overhead incurred by the mobile node as a participating peer in the overlay. in our Active Participation scenario. V/R) and then interpolate the instantaneous voltage readings MEASUREMENTS AND RESULTS In this section. For each experiment. we vary the size of the total overlay network size to observe its impact on local energy consumption. with each node running a complete instance of Chimera on a separate port.

8%) when the node was performing node join into a network of 200 nodes.4 watts. we estimate its CPU load as the percentage of time not spent on the idle process. We see three peaks in the histogram: one around 1. We summarize these values in Table 1. This is because the wireless card– device driver combination that we use seems unable to enable 802. The most significant histograms . Chimera generated the highest load (3. It is important to monitor CPU idleness to identify the CPU load and power energy (RtLkup) consumers and are route route row lookup lookup (RtRowLkup). The energy consumption increases only slightly when the card starts receiving data. We breakdown the energy consumption into detailed Overhead scenario. the expected energy level for power savings mode. Wireless Energy Consumption To characterize and better understand the energy cost of communication. Note that we do not see any peaks around 0. receive. with each function’s energy broken down into communication and computation costs. providing significant energy savings.55 watts. accounting approximately a quarter of all energy that is consumed by Chimera 2. Since Chimera is the only application running on Stargate. The idle task is a special task that the Linux scheduler switches to when no runnable processes exist in the task queue. Among all scenarios. Figure 7 shows the histogram of wireless card power con-sumption per time samples for one experiment (Control consumption that is due to Chimera. we analyze the traces that we collect for the wireless card. on average only utilizing 1-2% of the CPU time in our experiments. we measure the exact CPU load and energy consumption introduced by Chimera. By monitoring the idle task. one around 1. We produce for powereach components.11b power-saving mode in ad-hoc mode which we run our experiments consumption in.16 watts and one around 1. The results plotted in Figure 4 show that Chimera is extremely light-weight. with 100 nodes). this is because the RF radio must be on even during the idle state to listen for incoming packets.06 watts. Then the CPU stays in low-power mode until an external interrupt or a timer wakes it up. The peaks correspond to the maximum energy expended in the idle. and transmit states specified by the manufacturer.mode . where the columns show energy consumption of most CPU intensive functions. In Stargate. the idle task puts the CPU in a low power mode by turning off clocks to most of the CPU components.

Text-based MSN Client (TMSNC) [20] is an open-source MSN client for Linux. one to an x86 Linux desktop and the other to the Stargate. we installed two TMSNC client.55 + 1.16)/2) watts. one network thread and many worker threads. (1. significantly different in their software architecture.e. TMSNC initializes using login information and begins the . To produce this figure. Chimera has a multi-threaded structure. In this section. We believe there is one additional contributing factor. since it is only designed as a client.36 and 1. To evaluate TMSNC’s energy consumption. increasing node count from 50 to 100. with one application thread. has a much simple programming structure with only one thread that contains a single. And while not representative of all applications.11 watts. Client/Server versus Peer-to-Peer In addition to understanding Chimera’s own energy consumption patterns. We and also approximately 10% of all wireless energy is consumed during data transmission and 25% while receiving data. maintenance (such as heartbeat message and control messages) costs.experiment. we are somewhat surprised by the high amount of energy consumed for data reception. Given the low control traffic required by Chimera. We instrumented TMSNC’s source code to monitor initialization. The wireless power data shows that the receive energy consumption increases as we add more nodes to our network. we compare the send/receive modified each TMSNC node to generate chat messages using the same Pareto distribution as Chimera.36 (i. TMSNC and Chimera are consumption is between 1. It is light-weight and suitable for most Linuxbased handheld devices. large event handling loop. TMSNC. Overall. and break down the results into transmit and receive energy costs. and assume that the wireless card is sending packets if the power consumption is greater than 1. we take the histogram data. For example. we want to quantify the additional energy costs of running a peer-to-peer application compared to a similar application using a client-server model. increases the wireless receive energy consumption 5% in the Control Overhead scenario. comparing these simple applications should give us an idea of minimal energy costs introduced by the peer-to-peer model. Both applications are similar in functionality. and receiving packets if the power energy consumption of our Chimera-based Chat application to that of a client-server based chat application. We assumed the wireless card to be idle at all other times.

This is required because peer-to-peer nodes do not know when they will receive messages. Meanwhile. P2P ENERGY CONSERVATION We observe that a large percentage (> 64%) of energy consumed by the wireless interface is spent waiting for transmission during idle periods.com via https for authentication. The node would wake up after some preset sleep period and transmit buffered outgoing messages and receive incoming messages. . the peer-to-peer Chimera-chat uses only slightly more energy than the client-based TMSNC. Our measurements show that TMSNC generates an average CPU Load of 1.5 Joules in our setup). In addition. and allow the interface to “sleep” if no communication is received for some parametrized idle-time. There are some challenges that need to be addressed. While this initialization process is costly (5. While this ratio will likely change for applications that rely on servers for more computation. Since waking up the wireless card is not instantaneous. and plot its values along with those of Chimera for the same time period in Table 2. it does mean any inherent energy costs of maintaining a peer-to-peer overlay is minimal. In terms of wireless energy consumption. sleep periods that are too short may provide minimal energy savings. the length of each sleep session must be set carefully. As we can see. It then updates the status of contact list via the server. roughly the same as our Chimera clients. periods that are too long may significantly reduce achievable routing throughput. we perform simulations of nodes running Chimera-chat participation in each of We its run three each scenarios. We measured its energy consumption for an observation period of 100 seconds. In an adhoc network environment. A simple powersaving technique is to adaptively switch the wireless card to a low power state when no communication is observed. we find that TMSNC has lower cost.hotmail.session bycontacting messenger. An energyaware P2P protocol can monitor its incoming communications. Nodes can detect sleeping neighbors via missing acks and buffer messages for some finite period of time. neighboring nodes need to synchronize their wake-up and sleep time to provide enough overlap for some minimal data transfer rate. To estimate the potential energy savings possible using such anapproach.68%. and may cause neighbors to incorrectly assume the node has failed or left the area. when amortized over the session its overall impact is low.

This paper takes a first step by performing real energy measurements on a deployed protocol and application. context of Pastry. Kravets etal. Preliminary results show that it is feasible to deploy lightweight P2P protocols and applications on low-power.peer protocols from an energy-consumption perspective. we plot the average queuing delay added to each forwarded message. and plot the potential energy savings. as a long-term goal. We believe that by providing feedback from P2P protocol to physical layer.scenario with different values of idle-time. investigated wireless card energy consumption at both the physical and network layers. In addition. we can expand energy tests in scale. Finally. In addition. we are studying the problem of designing an energy efficient peer-to-peer protocol. HeteroPastry provided similar capability in the choose network coordinators. projects have examined how to adapt P2P protocols to efficiently utilize nodes with heterogeneous resources. while adding higher delay to forwarded messages. computationand including storage-. RELATED WORK While we are unaware of studies of overlay networks on energy constrained devices. embedded devices. including conserving energy by turning off redundant nodes in SPAN and a similar protocol called GAF. Both SPAN and GAF uses geographical information to bandwidth-intensive applications. Numerous protocols at the physical layer were suggested for ad-hoc networks. Finally. the PowerScopeprojectperformed power consumption measurements with setups similar to ours. there is significant potential for power savings if overlay messages can be batched and sent periodically. and also include more complex applications. Stemm et al. a shortidle timeout can result in significant power savings. we are considering the use . CONCLUSION To the best of our knowledge. In addition. As future work. showed that a similar policy in transport layer can provide 40% power savings for handhelds . Application layer wireless interface management can provide energy savings. Accordion allows different peers to choose its number of neighbors. it is possible to better manage network interface power states without any adverse affects to other applications that may be executing concurrently. thereby managing the tradeoff between higher maintenance costs and improved routing performance. prior work has not examined peerto. Clearly.

.of energy availability in neighbors in the construction of peer-to-peer routing tables.

.8.9 .7 /039.907.93 88902   %0 $9..3/0/ 5.030  3 98 80..08847    41 #  .8 .8:702039 2094/44 43 90$9.. 30.7-0 4:7 20.  390 $.. 70088 090.90.8 34 /85.8 98 4507. 809:5 ../08 /0-:3 538 ..3/ .90 574.943 8 4:7 .9002-0//0/.0  .  425.7.9:70 94 90 0099 !..0398 3 .3/ 35:9.7.8945 0.7..74 5407 803847 30947 .545:.7.90.7.$9.831..39.0 574.7   - 39071. /09.7/  %0 $9..0/ /08.90/0.3/%7.7.00307.. 3 .9002-0//0/-4.75943 41 4:7 05072039.943  0 .90 7:38 3: 0730    .09.7/ !"  %0 $9.7.  :7 050720398 :80.7/  $09:5.425:907 .943.7..3/ /08.90  843 3 :70   .91472  0 :80 90 $9.0 .

438:25943 :73 540757413 0:80430.89439071.90.7.4:95:9547989.3/9070088.9.7/ .700.0:80 942439479070307.0  %080 538 05480 90 5407 8:55 41 $9...

3 5407 8:55 ..90  .7488 .33041 90 " 94 8.3/ 90 $9.49.4330. 7088947 9.0 ..250 90 .9 8 .7.90/ 3 80708 94 90 2..

.0/:3/0. 10 0.0/:3 /0.55.3/ .43/87084:943    /0.3/ 7908 90 .. 0307.7480.93 88902 .843 92032.70 970..98/.02438  .0  0 .4393:4:8 4 0730 8.0 0 :80 .385..7039 94 90 :807 -7.8438 94 .94397.907 543907 8 97..0  %0 .943 4 -:1107 94 .59:70 0307 -0.55.55.//039107 ..7  3.3 /....3/908./8  $3..43909 41 90 .943 970. 5:75480 4507..8.90 4 97..7. .  243947 03/ 708098 90 !  5479 .93 88902  2.0 10  .9.47 41 4507.

574.

/8 9810 90/0. 7:3  0 :80 98 /7.8 . 05072039.3 39071.0/7.9 .07 94 70/70.07/85.98 .9089 73 ..890.09498-:1107 0390:80770.9  1907 0.52:4  ..4390398 3 $ 1472...

574.

3/ 90 8.9 90 7..0 .7941280.07  8 92089.55.943.7.0 108 94 ..0 ..0 . 10 147 41130 .7480.:/0 92089.90/0.4...52:4 394 .0/:07 .88 3.90/-90$9.425:90 0307 ...8 .040..70 0.07 0..55.943  8.:7.08 .3.0/:07 97.40.3/ 507 1:3.2580307.0412.090/.43/8   0 ..9094900.90 90 5407 97.9.94384.43/8 %080 94 97..55.0 108 -49 3.943  %0 .88  11303.3/ 5407 97.47/07 %0540797.438:25943 507 9..42-30 90 .943 97.8 2070/ 3 920 806:039.3/ 90 .3:..250/ :83 90 " .4770. .

90 . 8.943 2.59:70 574./8 -4.0  0./  /0.7-0/  90 " 8.:7 %0 8..47/ 90 .9 0 .748890! 54799483..:943  0 /0.0/:07 97.9 90 -.070.0398.8  %4 .843  0 70.49.:08 -09003 90 $9.0 43 90 7088947  %4 ..0/:3 /0./8 3 4:7 4.74:3/ 147 70:..8  039708  034: 94 .0/:3 /0.9.9438  0 300/ 94 34 03 970.2508 90 .438:25943 9 850.7430 92089.0730 97.. 8.55.0  %0 73 .3/ 1:3.4..:7.0/:07 /0. 970...084.07.8438 394 .425:90 90 909 9.90 2.0 94:9 39071073 9 4507.3  4.0 /7.59:70 90 8.4390989.-039077:590/47-4.1.25 .59:70 .3.3/03. 47 .../8 ..3/ 90 "  8 0 /08.07 70..4393:0 94 7:3 .7 2.07  %8 /0../0 :79072470  %0 970.0 /7.. 8....425:90 90 5407  0 1789 .03 0390889028949.3903.0//:7388902 .:943 00. 9 09 088 9..88  ..08 - :83 9707 0./08 98 31472.0/:07 97..47/8 .8438 147 4:7 2439473 5074/  3 0. 73 9.7.93 00.0450/ .9 0 .943  %4 .90 3 0730 85.08 9. :5 0307 .

.:77039  0  '.

5574.8:73 0307 050720398  0 .49. 70.70 0.998    .943  . . 0307 .7/ 8 2.07.935007390 4.11.90 .425430398  $0.4:8 7080. 8 5407 .9 43 4..90/ 94 4:7 $9..1907.$.9438  4:7 .03.438:25943 3 5407 8.3/43147../ 05072039  0 20..943.90 90 0307 4.5945 .77..59458 0.43/  0 .3/70.088     . 8 0307 .90/ 9 .748 147 90 !! 4. 34:7.07.:/08 9700 /1107039 . 4907 207..:770/ - 90 24-034/0.9 2088..3/   .3 0.1907 .33. 848 .5.7/ 89.8 3 97.943 8.998 3 97.0 94 479443.3/ 70.8.831.4330.03.. .0 24/0 .9438.2:2 5407 .55.7.438:207410307 3 .938..9.070.  :7 34/0 /408 349 39.070.. 39080.74 43 309478 41 80       .7.7.438:25943  0 7.34/07..03.03./ .039 807.38107 .08   %74:4:9 9080 ..94397.77.425:9078 .33 .425.8.  789  3 90 43974 ../ 4.07. 94:9.  0 9..9./943.5..90 .438:25943 7:333 .43/ !.39.3/2.438:259437.3829 ..943 057080394:720.3 .3/42.90 5479 4/0.7488 90 94 .3/   47 0. /0.74  9024-034/0.041 207.0.44808.90/ :83 90 !.08 ./38 $&#%$#$&%$ 39880.4898 0 433. 920 3907.. 0307.882:..08  !70. 34/0 8 ./ 3.7 90 80 41 90 949.90 34/0 0.7/0/.0.38 24/0 8  .07.998 3 70.#  .9.9:70/4.9 2088.39.97.11.90/ :83 90 !.3/ 903 390754..0 9 8 0307. 34/08 43 94 ...94341909209070088 .425090 389.74894979484.8:70 207.55...072502039.5.0 !..79.90 90 389. 34/0 7:333 .748 :7 20.438:25943 8..438:25943  789  0 20.8:702039 3.3.90/:8390 !48843 574. 05072039  0 7:3 .90/1107039 207.0 ..943 8..03.0!.0.. 805.8:702039 708:98 3 /09.08 .0  83.943 0307 ..70 4.070.50  8.79.. 8.389 .07. 24/0  4/08 .07.0 98 25.920   01789/8.90841 90 70088 39071.7094 /897-:943 94 24/03907 2088.3.0.0 89.4330. 30947 80 94 4-807. 4.39.5. 4.0397.88.7/897.908  .1 ..897:.8:70 207.34907 34/0 94 2088.47/3 94 850..03943207.30...9.7094 /897-:943 8..71:3.55.70.08   3...08 94 20.0803/8. 0307 .74  0 97 94 84. 147 4907 5007 4 .43.0/.79.03  9 0.:880307.438:25943 :3/07 /1107039 4507.3/0/ /0.422:3. .07.9438 3 .70 8570.0 . 70088   - .03.9. 570/0130/ 808843 920  0307.304:8.

9438  9 0.0/ .8 9.043:93 4190!&920 3 4:7 050720398  243 .74:3/ .38 243947390/09.0/7.2 430 .9 90 3: 8.74 9 34/08 0800 9700 50..55.03./43 90 0307 .3/97.90 90/09..438:20/ - 207.74:3/   ..3/ 90 0307 ./.748  207.798 70..425430398  070 90 .90 .3/0307 .7907 41 .0 1:3.438:25943 9. 8 90 43 . 24/0 ..438:25943 41 2489 !& 39038.9 147 90 70088 . 0307 9.9 0 .10/ -90 2.08 94 03 34 7:33. 8.7/ /0.943 8 0307 -7403 /43 394 .9070 .3/ -09907 :3/0789..03.0 207.7/ 89.943 7:333 43 $9.24/0 08:22.70. 6:.4.4898  %0 2489 831.438:25943 8947.3 09073.039.90/03070.90 94 89031473.-0 94 03.422:3.9 8 /:0 94 207.943 . 39077:59 47.438:25943 3.28 147 0.9439...9.3/ 5407 .:80 90 70088 .3/ 430 .39 0307 ./4 2:89 -0 43 0.943 .3 24/0 3 .9207.425430398 %0390!&89.3 /.39 03078..908850.9:707  %0 0307 ./    039034/0.0 .03 /:73 90 /0 89./3831.90  0 0892.0.  %0 /0 9. 0 7:3 4:7 050720398 3  0 574/:...7.0/:07 89.8  0 20. 9../ .:80 90 # 7.709080..07.3829 89.8.40.  98 8 -0./ 4.098 4909./ .070.7.4 540724/0-9:733411.9 207.998 ..438:25943 394 /09.998  90050.47708543/94 90 2.  $3.2:2 0307 0503/0/ 3 90 /0  70.8 6:0:0  3 $9.83 4 5407 24/0 :39 .8:70900.38 24/0  %8 8 -0..3/ .998 %050.90/4 349800.9 8 .-0   - 5407 8.3:1....489 41 . 1:3.4:393 .07.9!&4.425:9. 0307.0.42-3.422:3.943  0 ./8.438:25943 3974/:..350.74:3/   .90:8080028 :3..74:3/ .998  430 .  70088307438:25943 %4 ..4:238 84 0307 .8 3 90 8947.7.2508 147 430 05072039 43974 ..8 90 507. 8 097020 9 09  43 ..7/ 5407 .089:5 574.90 98 !& 4.08 9.-0  9825479.7/  :70  848 90 8947..4235.3.088 %0708:9854990/3:7084 9.3994243947!&/0308894 /0391 90 !& 4.438:2078 .08808 089 3 90 9.-0 574..3/ 74:90 74 44:5 #9#4:5  ..0 41 920 349 85039 43 90 /0 574.8 8 .85071472334/043 394.8. 850.43 8:25943 507 920 8.808 43 89 03 90 .0/ - 207.90/ 90 089 4..:083%.3094741 34/08 0-70. .2 41 70088 .0 5407 .894248941 90!&.014754078.0 90 97..55742.70 74:90 44:5 #9:5  .85:9890!&3.

4898  %4 574/:.9 90 70088 ..3829 .0981905407..0981:70 09.9079.9.3/-70.2 /.88:20 9.3   0      .0 0307 .  .0908947.3/ ./4390708:98394 97.438:259438 70.3/ 70.0.05072039 .7/ 8 803/35.

8089070088 70.808 . 8250 5747.39 /1107039 3 907 8419.3 /0.0/ 94 %$ .55.79-0.039 .039 430 94 .83 34/0 .7/94-0/0.// 2470 34/08 94 4:7 30947  47 0.039 807.80/ .943  2.8 0.70.2:9 970.0.9 41 .55. 41 232.9 3/ 0349705708039.0 .70.70..3 4707 970.55.223 897:.3.9.7.8.55.438:25943 3.4/0 94 243947 39.3 5. 0307 ./ .3/ 2088.03.0 :8 .3/ 8:9.4898 3974/:. 2:.943.74  .08  %$ .8 ./  430 30947 970.43974 2088.9 .80/ .9:70  9 430 .3/ 207. .8 0 .438:25943 0 389..90  0 3897:20390/ %$ 8 84:7.7082.425.08  .38 .9 .3/0/ /0.0391473: 989 09. .70 0.9 .8 .731:3.3 4503 84:7..439. .998  0.0 8:.438:259438-09003 .0 0307 .9438 84:/ .9438  .55.438:25943  3 90 43974 .041. 848 9..9 90 70.4:391742 94 3.80/ .943  %09 -.70 .039  ..438:25943 41 4:7 207.0 803/.943 94 9.0.3/ . .07 -.9:70 9 43 430 970.9438. 4907 9208  %0 70088 5407 /.0 9 8 43 /0830/ .0/ - 90 5007 94 5007 24/0  %4 0.3/ 2.3/9049079490$9.943 970...9./ 9.0.70 831.3903.90.7.73 9080 8250 ..3/ .9:70 207./ 8.  0307 ./8  %$  83. -./0/ 897:..070.098 1 90 5407 ...88:20/9070088.3/3 445  49 .. .9 .-0 147 2489 3:-.250  3...90%$ 80307.07.9 .55.3  3:/08945.3/ 70.80/ $ 039 %$   ( 8 .0 $.998  ..:. 830  .0 0307 .

03 90 4 .943 .  .//943.55742.84 24/10/ 0.3828843.947  039.3/ 0 70.0.24:39 41 0307 .05943  0 -00.9.9 2088.0 .0.  0 .3/ -038 90 .90 ..70 8420.20 !.43974 97.438:20/ 147 /. %$ 34/0 94 0307.70.4397-:93 1.11.9.90 41..  %$ 39.438:20//:73/.9. 70.7008803078 . .08 :83 43 31472.97.3 /..9 8:75780/ - 90  .4898  0 .0 9070 8 430 .8 207.7094 /897-:943 . 706:70/ - 207.08 :83 90 8.

039 807.39 94 6:.7.07.70/ 94.943..5007 94 5007.55.//943.//94394:3/0789.425.990738  0 .70 90 .943  0 .07 24/0  3 98 80.$07.078:8!007 94 !007 3..3/3207. 843 0307 .55.4898 41 7:333.438:25943 5.425.82.943:83..391 90 . 0307 .

9 89 .8 407 .20. 831.748  0 7:3 0...0 89.3..81.30 5074/89.9. .30 147.0.385488-0:83 8:.08 .3/97.  !!# $#'%  04-807..9 300/ 94 -0 .43 9 9480 41 207.39 70/:.0.0 8 85039 .943  9 903 :5/.943 9/40820.9 %$ .3 243947 98 3.3 4-807.0 !& 4.70944432.//70880/  3 .039 -.094 8005 134.9320880307 492..0398 3 90728 41 70088 0307 . 574.7.59.908 . 8005 808843 2:89 -0 809 ...3/ 70.25..3.07.0 :5.39.3 .70 8420 . .9438  .0/14784205..-0 74:93 974:5:9  .3/ ..43/8 ...5..8:70/ 98 0307 .943 8..-0 80.088 8 .. 0307 8..9003 34 . 0 5071472 82:.7432039  30-473 34/08 300/ 94 83. 90 807.9438 41 34/08 7:333 207.1907 8420 570809 80055074/.422:3.9 3 0.70 944 8479 2.0/47019 90 .9. 0307 8.438:25943 147 .8232.7/ 8 349 389.3/549 98 .039.3307039 0307 .88:209034/0. 2883 . .36:0 894.. 97.07..9 .489   4:08 3 4:7 809:5  03 ..808843-.0308 9..9.423 2088.489  0 20.07..38107 7./0 232..90 70088.3/800592094 574.0 3.0 .09. 5007 94 5007 4.0/  3 0307 .9:80843 89247003079.425:9.38  0.90 90 549039.93 147 97.20970//0 920  %0 34/0 4:/ .70.80/ %$ 0987.3 800 905007 94 5007207.943 870.07  0 98 39.//943  90 039 41 0.9 %$ 0307.5 147 8420 232.08  4/08 .3829-:11070/4:943 2088.90  3 ..7430907..0..0781472470 .701:  $3.07.438:25943  0 13/ 9.79..4 90 39071.:80 30-478 94 3./4. /.0. 30947 03.9435074/41  80.33 .84:7207.422:3.943 8 4-807./ 41    74:908.943 574..304:8  8005 5074/8 9.0 2088..4898 41 2.390.4770.0 .940.0790808843984.:80 5007 94 5007 34/08 /4 349 34 03 90  70.55.42 ..4540789.438:20/ - 90 70088 39071..03./.08 147 8420 13905074/41920   %070 .7/94.24790/ 4.423 ..0   41 0307 ..9 80053 30-478 .984  :7 20.97043807..3828843 /:73/05074/8 %88706:70/-0.908 90 89.9:8 41 .94389...439.  %4 0892.. 9958 147 ...3 /090.5574.:08 . 41 98 9700 5... 147 908.422:3.:9039.3/ 2.9.390.3/ -:1107 2088. ..70 !! 57494. .209205074/3%..8:7020398 84 9.08   8250 5407 8.3 .3 :5 90 70088 .0 :5 .8 .4 .70507...39./0 034: 4.439.

9478  3.07..:77039 3.0780 .59 !! 57494.55.0 30947 39071...03..438:25943 .0 5407 89.08  ..5.7 .3.3/./.07 .0 ..7 54.. .   3.4 94 58.3 -0 -.110.38  3 .3 574..74 9 /1107039 .3/030790898 3 8. 43 9072 4.55..7708:98 84 9.08 .:08 41 /0 920  .55..  0 . 1789 8905 - 50714723 70.3/0/8  :2074:8 57494.943..438:25943 507850..943.438:25943 20.47/43 .447/3.908 94:9 .. 309478 43 0307 .06:0:3/0.3.8 349 0.4480 30947 .:93. 1742 !! 57494.07 985488-094-099072.3/ 549 90 549039.7944:78   &$  %4 90 -089 41 4:7 340/0  5747 47 .. 309478  3..0 0.897  574. 31472./3 100/-..9 9 8 10.3 0307 - 9:73341170/:3/.08  #% # 00. 3 97.3.4./ 4.//0/940.8.230/ 4 94 .48 .0 0.-000.0  09074 !.7 .3/  :808 047.//943  9070 8 831.07.38 1 4.48 /1107039 50078 94 .3903.438/073 90 :80 .07.7/0/2088.30/ /0.3.38 147 .38 0.3//9 39038.43897....039 5007 94 5007 57494..94 147.3/ .3 0307 011.4505740.  840/ 9..9 .98 94 4907 .:/0 2470 .9438 9.0  .43.0  -.4250 ..70 .943 !7023. .230/ 500794  5007 57494.70:3.98 .39 54078.0770088 39071. 82.02.9.  90 !407$.3 .3934/08 3$! .425:9.08  5740.9..089.48 94 011.4 .9 -49 90 58...07 070 8:0890/ 147 ./0411 -09003 07 2. 147 5407 8.3/ 2574. /0540/ 57494.38  $9022 09 ../0   5407 8.3..8-0 94 /054 9 09 !! 57494.84 3..8.48 .0 9.//307/0.3/ 30947 .704189:/08414.9 - 574.3 708:9 3 831.55.0 .//943 054990./00307 8.4  0 -00.  81:9:7047 0.//943 0.9438 43 4 5407  02-0//0/ /0.9438  .385479 . ..7089:/390 574-02 41 /0833 .3574.7/ 0307 .48 1742 ..8:7020398 43 .147.:/3 8947.08  3 ..0/ .9438 3.:/3 ..90/ 70088 .943 94 .55.. 0307 8. .897 55.4390941!.039 :90 34/08 9 090740304:8 7084:7.3 90 97.39 549039.7/0/2088..8:7020398 9 809:58 82. 82.. 8479/0 9204:9 .3/..3/80395074/.7 57494. 0307 20.4898 ..3/ 3..950714720/ 5407 .305.3/ . 2088././0/ 82.08 .4480 98 3:2-07 41 30-478  9070- 2..0  %8 5.02039.098 09.9 90 58.507 9.078  7.43807.0/ 74:93 5071472.0/   49 $! ..5.-9 3 90 .3 0307 .9 2.

.-08  .41 0307 .943415007 94 500774:939.43897:.-9 3 30-478 3 90 ...