Master s Thesis 修士論文 論文題目 CACHE CONSISTENCY IN ICN: LEASE 5114FG21-6 THEINT THEINT MYO. Hidenori NAKAZATO. Supervisor 指導教員 年 7 月 1 9 日

Transcription

1 Graduate School of Fundamental Science and Engineering Master s Thesis 修士論文 論文題目 CACHE CONSISTENCY IN ICN: LEASE Student ID 学籍番号 5114FG21-6 Name 氏名 THEINT THEINT MYO Supervisor 指導教員 Hidenori NAKAZATO 印 年 7 月 1 9 日 WASEDEA UNIVERSITY i

2 INTERNATIONAL PROGRAM FOR SCIENCE AND ENGINEERING CACHE CONSISTENCY IN ICN: LEASE BY THEINT THEINT MYO 5114FG21-6 A THESIS SUBMITTED TO THE SCHOOL OF INTERNATIONAL PROGRAM FOR SCIENCE AND ENGINEERING WASEDEA UNIVERSITY IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE (ICT) WASEDA UNIVERSITY SEPTEMBER, 2016 提出日 2016 年 7 月 19 日 ii

3 ACKNOWLEDGEMENT The author would like to express to her special thanks to her academic Supervisor, Professor Hidenori NAKAZATO, the Graduate School of Fundamental Science and Engineering (FSE) for his valuable advice, for his kindness, care suggestion for the research study, for suggestion for the idea of research, for giving guidance to the presentation, the preparation of the thesis and for reviewing this manuscript. The author would like to thank to lab s member, who discuss in lab s seminar, for their suggestions about the research. The author would like to thank specially to the members of the board of examiners, for their guidance, helpful discussions, useful comments and suggestions. Greatly thanks to all professors for their valuable lectures and office staffs working at the Graduate School of Fundamental Science and Engineering (FSE), Waseda University, for their supports and helps. Many thanks go to the organization of Japanese Grant Aid for Human Resource Development Scholarship (JDS) and JICE offices from Myanmar and Japan and also responsible persons of Japan government officials giving her a chance for this master course. Finally, the author would like to express her thanks to her parents and family members for giving her encouragement all the time to finish this course. i

4 ABSTRACT Information-Centric Networking (ICN) is the next generation network technology. In ICN, cache consistency is important for some applications like disaster application where content must not be stale data in providing to the client for important information. In Disaster, information always has to be updated accordingly to the condition and client should receive the updated information when client request. So, in Disaster application, cache consistency is a factor to improve the performance of application. In this paper, Lease mechanism will be used to get the cache consistency with the higher performances like low invalidation message, low delay time while updating and low network load. We propose some mechanisms which are added in original lease to get the higher performance while eliminating the weakness of original lease. Lease is a time-based mechanism to provide the cache consistency. But it have still some weakness like need to send message between server and cache before updating the contents and if one of cache cannot send back the acknowledge message to server for updating, there will be delay time for updating the data. So, we propose some technique that is to adjust between the update time, cache time and the replacement time to overcome some of the above weakness. We simulated in ndnsim simulator and the number of control message between the server and the cache and delay time for updating content are presented as an evaluation results. According to the simulation results, propose mechanism has better performance than the original Lease. ii

5 TABLE OF CONTENTS ACKNOWLEDGEMENT... i ABSTRACT... ii TABLE OF CONTENTS...iii LIST OF FIGURES... v LIST OF TABLES... vi Chapter INTRODUCTION Introduction to ICN Motivation and Aim of the Research Thesis Organization... 3 Chapter CONTENT-CENTRIC NETWORKING CCN Overview Basic Operation of CCN Basic Functions and Features of CCN Other Features of CCN Transport Layer of CCN Flow Control, Sequencing and Naming Challenges of CCN Naming Caching Routing Security Chapter Caching Caching Policy Cache Replacement Policy Cache Consistency polling Every Time iii

6 3.2.2 TTL Lease Proposed Scheme General Operation of Proposed Lease Scheme Calculation of Expire Time in Producer Updating in the Producer Replacement Policy for Proposed Lease Scheme Chapter CACHE CONSISTENCY: LEASE Introduction for simulation Simulation Challenges to provide content with tag Content Specification in Producer Simulation Challenges to store in Content Store Simulation Challenges of Discard entry and Replace the cache Algorithm for Overall Operation Evaluation Results for the Proposed Lease Chapter CONCLUSTION AND FUTURE WORKS REFERENCE iv

7 LIST OF FIGURES 2.1 CCN Packet Types CCN forwarding engine model CCN Overview Example of Data s name Content Base Addressing Association of Trust establishment and the publisher Keys Caching about Cache All Caching about Cache Probabilistically Caching about Cache Selectively Caching in router with the higher probability to cache LFU replacement policy FIFO replacement policy Operation of Polling Every Time Operation of Time To Live Normal Case of Lease Updating and delay happened in Lease General operation of Proposed Lease Scheme Procedures that the server operates to provide content with expire time Producer and the net device face how to operate for Traditional Lease To store expire time in CS, on Receiving Data To store expire time in CS, on Outgoing Data Producers to discard the CS entry whose content is expire Producers to replace the CS entry Algorithm for Overall Operation 37 v

8 LIST OF TABLES 4.1 Content Name with Lifetime vi

9 Chapter 1 INTRODUCTION 1.1. Introduction to ICN Nowadays, many researchers are working on Information Centric Networking (ICN) because it has many advantages over the IP network. The internet can provide can provide a connection between two people, who are far away, by phone or video chat and also provide some source of data and knowledge not only to students but also to other ordinary people. With the most advanced technologies, people can control and check their facilities at home while they are outside. People use many network devices like phone, computer, tablets and other devices for their work, business and also for entertainment in their daily lives. So, the usage of internet is also growth. Therefore, there are many requirements to provide the better quality with the internet services, like Bandwidth, Throughput and the delay time. The protocol used in internet is IP Technologies and nowadays, all are worry for about the lack of IP, it will be run out one day. So many researchers are trying the advanced technologies ICN to replace the IP to get best performance of internet like mobility, security, throughput and requirement of bandwidth. The purpose of many proposals about ICN architecture is to solve the above problems. In general IP network, if the client request content like video, movie, data or web page, the request will go to server and reply back through the network. If many client request same contents, these many requests will go to server and server have to reply all client through the network. As a result, there may be heavy traffic, more bandwidth is required and most of the components in network have to work almost all the time and this will have huge energy consumption. In IP network, there are many limitations when the request or the number of users 1

10 becomes higher. In ICN, content can be stored in routers and can be provided when another user requests it if the cache is valid. So the usage of bandwidth can be reduced according to the cache, round-trip time is also short and some of the devices which are located outside of the path between clients and the cache can be in sleep mode for a while to save energy consumption. ICN is the potential technology to replace the IP network Motivation and Aim of the Research In CCN [1], user sends the message for requesting content with a specific name through the network equipment to the one who holds the requested content which can be producer or routers. Contents can be cached at any routers in the network and also more than one place. In IP networks, mostly only servers can provide contents and not the network equipment like routers. In ICN, not only producers but also routers can cache contents and if the content name specified in a request message, called interest packet, matches with the cached content, the cache will provide the content instead of its producer. In this case, the round-trip time for the content can be reduced and traffic in the network can also be reduced. But consumers can get stale data from this cached. Stale data means that the content at its producer is updated while the cached content is not updated. The cached content will keep providing its old content. That makes consumers to receive stale data. So cache also needs to be updated after the content in the producer is updated. This means cache consistency. There are many mechanisms to keep cache consistency, polling every time, TTL and Lease [6]. We can use these mechanisms depending on the applications. 2

11 In this research, we emphasize on a lease. With Lease, when consumer requests content, the interest message requesting the content is forwarded to its producers through the network and the producer will provide the content back to the consumer. While the content is forwarded to the consumer, the content will be cached at one or more places in the network. Content can stay in cache according to its lifetime. If the cache expires and another consumer requests again this content, the interest package will be forwarded to the producer again and the producer will provide. But in Lease, there are some problems. When the content in the producer is updated, the producer has to send the message about the updating to all caches whose contents are still valid and all caches have to acknowledge to the producer. If one of the caches can t reply, the producer has to wait till the cache is expired. So there will be delay time in updating data. Therefore, Lease has strong cache consistency but it has weakness like delay time and overhead messages between the producer and the caches in updating the contents. In this research, we propose some techniques to overcome the Lease s weakness, delay time and the overhead messages in updating the contents Thesis Organization The following chapter 2 is about the overview of basic CCN data structures, packet forwarding operation and the cache. In chapter 3, we discuss about caching policies, cache replacement policies, cache consistency scheme and our proposed Lease mechanism. In chapter 4, we describe about the simulation challenges and evaluation results. In chapter 5, we summarize our thesis and the future works. 3

12 Chapter 2 CONTENT-CENTRIC NETWORKING 2.1. CCN Overview CCN is the Content Centric Networking which based on the named data communications architecture. It is the next generation of IP network with the advantages of strategy, security and so on. Its routing scheme depends on the name of contents instead of locations [1]. There are two types of packets in CCN which are Interest and Data packet as shown in fig 2.1. Interest is the request from the consumer and it contains name of requested data (content name), selector which is for discovering and selecting the best data that the consumer wants and nonce to detect looping interest. Data packet contains the data that the client requested, content name that will be satisfied if it is the prefix of the content name in interest packet, signature, signed info and Data [1]. Figure 2.1 CCN Packet Types 4

13 Basic Operation of CCN In IP network, when the request message from client reaches to the router, router will check the destination IP address and make the routing accordingly to this destination IP address. When it reaches to the source, source will send back the requested data to the location which is corresponding to the source address in the packet [1]. In CCN, there are many differences with IP network, like strategy, efficiency of memory usage and so on. When the consumer requests the content, the interest packet will be forwarded to the producer which may be main source or cache router. In CCN, there are three types of data structures. They are Content Store (CS), Pending Interest Table (PIT) and Forwarding Information Base (FIB) [1]. Content Store (CS) is like the buffer memory of IP router but it can replace the contents with the different replacement policies like LRU, LFU and FIFO replacement policies. CS can store the data packet for other consumers who will request in the future. And this is called cache [1]. Pending Interest Table maintains the condition of the interest packet that forwarded to the producers and records the data name that contained in the interest and faces. Each PIT entries are used when the return data is sent to the consumer [1]. Forwarding Information Base is like the routing table and is used to forward the interest packet to the producer which holds the data name same with the requested one [1] Basic Functions and features of CCN When the consumer sends the request, called the interest package, will be arrived on face. A longest-match lookup is done by checking content name in interest packet. 5

14 Accordingly to the CCN forwarding engine model as shown in fig 2.3, Content Store match will be preferred over a PIT match and over a FIB match. If the requested data packet is already cached on Content Store, cached data will be sent out to the face that the interest arrived. And interest will be discarded. If there is no in Content Store, PIT match will be done If it match, arrival face will be added to the PIT entry and the interest will be discarded because the same interest already forwarded before this. After the producer sends back the data packet, a copy of it will be sent to the face that the interest arrived [1]. If there is no match in the CS and also in the PIT, check matching of FIB entry and there is one, interest will be sent to the producer. And then new PIT entry is created from the interest and its arrival face [1]. Prefix FIB Face List /waseda.jp/ 2,3 PIT Prefix Requesting Face /waseda.jp/naka_lab/index-e.html/v2/s1 3 Content Store Name Data /waseda.jp/naka_lab/index-e.html/v2/s1... F P C Ptr Type Index Face 3 Face 2 Face 1 Application Figure 2.2 CCN forwarding engine model As an overview basic function of CCN, as shown in fig 2.3, when the consumer sends the interest, this interest will be forwarded to reach the producer 6

15 because there is no cache in the first stage. The producer will send back the content to the requester according to the forwarding strategy. While forwarding, one or more routers can cache the copy of this content and save in the content store for the next request for this content. When another consumer requests the same content, the interest will not be forwarded to the producer because the content is already cached in router and it can be forwarded to the consumer instead of the producer. Data 1 Interest 1 R3 R2 Cache Data 1 R1 Server Consumer 1 R4 Interest 2 Data 1 Data Interest Consumer 1 Figure 2.3 CCN overview 2.2. Other Features of CCN Transport Layer of CCN CCN transport layer operates on top of unreliable packet delivery services. So the interests and contents may be lost when they are forwarding. To overcome this, interest should be retransmitted. This is done by receiver s strategy layer [1]. 7

16 Data in CCN cannot be loop because of the discarding of duplicate packets by the basic node mechanisms. But the interests can loop. For this problem, interest packets include a random nonce value to check and discard the duplicate interest [1] Flow Control, Sequencing and Naming In IP network, congestion can occur in the links between routers and it affects the delay, packet loss and disconnection of network links. To eliminate this congestion problem, senders have to adjust their window sizes dynamically. And also some protocol like Spanning Tree Protocol can be used [1]. But in CCN, flow balance is maintained at each hop and no additional techniques are needed to control the congestion in the link [1]. Data is identified by the simple sequence numbers in CCN. Consumer always requests individual pieces from large collections of data, so CCN need sophisticated techniques [1]. User/App supplied name Versioning and segmentation Human Readable: Binary Encoding: /parc.com/photo/supera.jpg/_v<timestamp>/_s1 Globally routable name Organizational name Conventional/automatic 8 Parc.com 6 photo 11 SuperA.jpg 7 FD04A Figure 2.4 Example of Data s name 8

17 The difference of CCN network with the IP network is that CCN based on the name as mentioned in chapter 1. Name is constructed in hierarchically so that an individual name composed of a number of arbitrary octets that shown in fig 2.4. Name notation used URIs with / character separating components. But the delimiters are not included in encoding and in name [1]. Hierarchically structure is as shown in fig 2.5. The request name is /parc.com/photo/supera.jpg/_v<timestamp>/_s1 so first in Parc.com and then choose photo and the addressing of name is in highlight. The requested name is the photo with the first segment of version 2 accordingly to the example [1]. Parc.com photo SuperA.jpg _v1 _v2 _v3 _s0 _s1 _s2 Figure 2.5 Content Base Addressing 9

18 2.3 Challenges of CCN CCN is the next generation of IP network and it has many advance techniques over the IP so many challenges also meet in implantation of CCN. Most of these challenges are Naming, Caching, Routing and Security described in detail as in following Naming As mention in section 2.2.2, CCN is based on the name data network and its structure is hierarchical. When the network is huge, naming of each content is also becoming more complex and the matching also takes more time than the normal network Caching One of CCN feature is caching and it objective is to reduce the traffic load to the server, to store content as much as possible. With the caching, the loading of content store and required memory also have to count in challenges Routing CCN name have no limitation for its length so Routing will also take more time for this name. And Routing Table size also needs to control and update for routing information Security CCN is built on the notion of content-based security [1]. All content in CCN is authenticated with digital signatures and private content is encrypted. 10

19 In content validation, CCN authenticates the binding between names and content, the signature in content that mentioned in fig 2.1 that is over the name, content, and signature verification, signed info. This allows content publishers to bind securely arbitrary names to content. A name-content binding was signed by a particular key. Each signed CCN Data packet contains information to extract public key as a shorthand identifier for the publisher. How the trust establishment associate with the publisher keys is illustrated as in fig 2.6. /parc.com/allie/videos/widgeta.mpg/v2/s1/0x4edb96a4... Signed checksum 0x1c Key Parc.com/allie/desktop public key Signed by Parc.com/allie Signed by Parc.com Figure 2.6 Association of Trust establishment and the publisher keys 11

20 Chapter 3 CACHEING IN CONTENT CENTRIC NETWORKING 3.1. Caching CCN is to overcome the data traffic explosion problem caused by a large number of file sharing and video streaming among many users by providing the many advantages over IP network. Caching is also included in these advantages. Caching is the kind of storing the data in memory and used for further use. As in general, when the consumer is request the data, called the interest packet will be forwarded until the data is found. When the content is found, it will send back the content called the data packet to reach to the consumer. In this forwarding, one of the routers on the path will be cache the copy of this data that is store in content store of router. If another consumer requests the same data, interest will be also forwarded until the content is found. In this case, the previous router will be provided content from its cache. In this way, interest no need to go to the main server. The benefits of this caching is that main server traffic load, bandwidth usage can be reduced and round trip time also can be reduced that means consumer can get the content faster than from the main server. When the requested data is found in one of the routers, it means cache hit and if no cache and interest packet have to go to main server, it means cache miss. Cache hit is usually measured as a cache hit ratio or probability of hit. This is depend on several factors like the cache size, number of request, number of content and the network size that is relationship with the number of router [2]. To store at content store, mostly frequency request content or more popular content are decided to cache depending to their popularity, requirement of memory 12

21 for caching and kind of caching scheme. There are many kinds of cache policies. These will be described in following section. For cache to be stay as long as possible, to use the cache data effectively and the requirement of memory to cache the content, there are so many kinds of replacement policies in CCN that will be described in the following section also Caching Policy There are so many kinds of cache policy that implemented by other researchers. Mostly cache policies are based on three kind of caching for what to cache. These are Cache All, Cache Probabilistically, and Cache Selectively [3]. Another decision is where to cache [3]. Cache All When the content is send back from the server or cache intermediate routers, one of the routers that all the content through it will be cached without any checking the feature of content as show in fig 3.1. Data Packet 6 Data Packet 5 Data Packet 2 Data Packet 1 Data Packet 5 Data Packet 4 Data Packet 2 Data Packet 3 Data Packet 1 Cached data packet Data Packet 4 Data Packet 2 Data Packet 3 Data Packet 1 Data Packet 4 Data Packet 1 Data Packet 3 Data Packet 3 Figure 3.1 Caching about Cache All 13

22 In general, when the consumer request content, it will send the interest packet to the source and router that will not cache for this content will forward this interest packet until it meet with the data packet. When main server or one of the cached intermediate routers hold this data packet, it will send back this data packet to the consumer and other intermediate routers also have a change to cache this data packet. Whether cache or not is decided depending on the policies used. Now Cache All means all of the content that through it will be cached. Actually all of the content will not be cache in practical and to do so, requirement of cache memory will be higher and it will cost more. In other hand, latency for all request will be short than the other. Cache Probabilistically Data Packet 6 Data Packet 5 Data Packet 2 Data Packet 1 with P( ) Cached data packet Data Packet 4 With P( ) Data Packet 1 with P( ) Cached data packet Data Packet 4 With P( ) Data Packet 1 with P( ) Data Packet 4 With P( ) Data Packet 1 with P( ) Data Packet 3 Figure 3.2 Caching about Cache Probabilistically It is one of the caches that have one condition between the content store and the data packet depending on the requirement. This is like the filter between the content store and the data packet. If the data packet has probability that is met with 14

23 the filter requirement, this data packet is acceptable and will be cache by the router and also forwarded to reach to the consumer. If not, this data packet will not be cache and only forward to the consumer. In this way, cache probabilistically can reduce the number of read/write operations in the CS than the cache all that described in above. The main objective of this is to reduce the entries to the CS and to disable the caching of unpopular content in the CS while maintaining the storage of popular entries in the CS. This will make the cache hit rate will be higher. These techniques are illustrated in above fig 3.2. In figure, data packet with P( ) means that this data packet is match with the filter about the probability. So only the data packet that match with the required probability will be cache in CS and forwarded to the consumer. If the data packet is not match with the required probability will not be cache and will be forwarded only. Cache Selectively Data Packet 6 Data Packet 5 Data Packet 2 Data Packet 1 with C( ) Cached data packet Data Packet 4 With C( ) Data Packet 1 with C( ) Cached data packet Data Packet 4 With C( ) Data Packet 1 with C( ) Data Packet 4 With C( ) Data Packet 1 with C( ) Data Packet 3 Figure 3.3 Caching about Cache Selectively 15

24 It is similar with the cache probabilistically, but one case is that here selectively cache may be depending on the probability or other that will be some conditions. If developer wants to cache the videos, before cache to CS, have to check the data packet that is videos or not. Like this filtering will happen depending on the requirement criteria and can vary on demand. And also the content may be popular or unpopular. In fig 3.3, data packet with C( ) means that this data packet is match with the developer defined criteria and want to cache it. Only data packet is not match with criteria and do not want to cache. So the router will cache the data packet with C( ) as shown in fig. Where to Cache It is also one of decision for caching. Where to cache is that which routers will be cache the content. There can be divided into two parts that are all of the routers along the path will be cache. There is no additional feature to choose which router will cache it. And another one is only the specific router will be cache the contents along the path. It is just like the router which has a highest probability to cache the content can cache this content. This can be implemented by calculation the router s highest probability before cache and compare one predefine value and it the condition is true, this router will cache this content that is illustrated with fig 3.4 [3]. In fig 3.4, router with HP means that the higher probability to cache the content and it will cache the content that through it. One of the routers with HP cached data packet 1 because only data packet 1 through it and another router cached data packet 2 only this data packet through it. The router near the consumer cached data packet 1 and 2 because these two data packet through it [3]. 16

25 Router with HP Data Packet 1 Router with HP Data Packet 2 Router with HP Data Packet 2 Data Packet 1 Figure 3.4 Caching in router with the higher probability to cache Cache Replacement Policy Although the caching have many benefit to the network by reducing latency, network load in server and bandwidth usage, some of requirement are still needed to use the caching effectively. To cache the content, memory in router is required. If the caching content is higher, requirement of money also will be higher and entries to the CS, CS lookup and updating will be also considerable. If all of the router memory, cache size of router to cache the new contents are full and there are also still content that have to cache, some of cached content have to be discarded and replaced by new content to be effectively used the caching of CCN. So to be balance with these requirements, some strategies are requirement to be implemented. That is cache replacement policies. There are many replacement policies. Among then some are described in as follow [3]. 17

26 LFU Replacement Policy LFU stands for Least Frequently Used Replacement policy. The cached contents in the CS these are used least frequently will be discarded in first option. So to know the cached content is least frequently used or not, some factors are required to check it. That is to count the time of usage or access to that content as shown in fig 3.5. Router Data Packet 5 Data Packet Access Counters Cached Data Data Packet 3 Data Packet To replace cached content Data Packet 1 19 Figure 3.5 LFU replacement policy In fig 3.5, that show how the LFU replacement policy works. The number beside the data packet of cached data is the access counters of how frequently used of that data. By checking the value of this access counters, replacement or not can be decided. In this fig, Data packet 3 has the least access counter that is 9 so that this data packet 3 will be discarded and new content will be cache. In this way, cache memory can be effectively used. 18

27 FIFO Replacement Policy FIFO stands for the first in first out replacement policy that means the oldest entry of CS is replaced. The first entry in CS will be discarded when the CS need the space to save the new content [4]. Router 5 th Data Packet 4 th Data Packet Cached Data 3 rd Data Packet 2 nd Data Packet 1 st Data Packet To replace cached content Figure 3.6 FIFO replacement policy Fig 3.6 shows about the FIFO replacement policy. 1 st Data packet means the first entry (input) of CS so that it will be discarded to cache new content for replacement. And after that, when the next new content is required to cache, 2 nd Data Packet will be discarded and it will be continuous sequentially [4]. LRU Replacement Policy LRU stands for the Least Recently Used Replacement Policy and that are used in this research. When the cache size is full and other new content have to be cache, LRU will discard the least recently used cache from the CS accordingly to its policies. So active cache will be stay in CS. Active means most frequently request content. 19

28 LRU simple to use and also operation is fast because the running time for request is O(1) [5] Cache Consistency In this section, we will describe about cache consistency. As mention in about section, cache is that the store of content depends on the cache policies used to provide it without going to main server when the same or other consumer send the request for this same content [6]. Cache consistency is also a part of caching but they are a little different in objective. Cache consistency means that the main server s or source s content and the cache content have a connection and there is almost no stale data. Stale data means that the old content of cached [6]. When the consumer request the content, interest will be forwarded until the content is met that may be from cached content or main server. In general caching, when the server updates the content, server has the new version of content but the all of the cached data can t be new version of this content. Only the content the replaced by other content and consumer request again it and then content come back on its path that this router exits. This router can cache again the new version of it when the server sends it back for consumer s request. All of the caches can be updated depending on the consumer request after the main server updated the content. Traditional cache consistency depends on the main server because if the main sever can t update, there will be delay that will be longer depending on the cached content s lifetime [6]. But in some applications like the information have to be updated in time or on time, cache consistency is very critical to grantee for not stale data to the consumer. 20

29 There are also many techniques to provide cache consistency. Among them Polling Every Time, TTL and Lease will be described [6] Pooling Every Time[6] Pooling Every Time have strong consistency scheme [6]. But it has some considerable factors that are network load and the bandwidth usage. The operation of it is show in fig B 4 C A D Data Interest If modified message 304 Figure 3.7 Operation of Pooling Every Time As shown in fig 3.8, when the consumer requests content, in the first stage there is no cache. So the interest will be forwarded to the main sever and sever will send back the content to the consumer. Along the path, routers will be cache as the normal condition. But another consumer request again this content (1 in figure) and that is sure there is cached content along the path but cache first have to contact to the 21

30 sever to make sure that the cached copy is validated or not before sending to the requested consumer (2 in figure). This validation message is if modified since request. If the server replies with 304, it means data is not modified. So that cache can send its cached content to the consumer (3 in figure). If the server replies with the content (C in figure), cache have to remove its cached content and copy and cache the new content that sever sent. And cache will forward it to the consumer (D in figure) [6]. Accordingly to the operation of polling Every Time, in every cache hit, cache has to send message to server and vice visa. When the content is not modified, sever only have to send 304 so only at that case the bandwidth usage will be small comparing with the bandwidth usage in normal content sending by server. But in overall operation, It has strong cache consistency and a very small amount of chance to get the stale data by consumers that is almost no stale data. In the other hand, traffic between cache and sever will be still huge and also bandwidth usage also will be bigger. Latency also will be long [6] TTL TTL stands for Time To Live. TTL is like the life time of content that is predetermined. Content s lifetime is valid by its TTL value. For e.g, TTL of content 1 is 1 hours is that this content will be discard after 1 hours of its current caching time. So after the TTL is invalid and other consumer request for this content, there may be request hits on an expired cached. In this condition, router will send if modifiedsince request to the main server. Server will provide the content and router will cache copy and forward it to the consumer. This is shown in fig 3.8 [14]. 22

31 By adjusting the TTL of content, traffic between cache and the server and bandwidth usage will vary. If TTL is longer, traffic between cache and the server and the bandwidth usage will be small. If TTL is shorter and consumer request many time to the same content, interest are needed to forward to main server frequently and server have to provide content. So that traffic and the bandwidth will be higher but for the consumer have small chance to get steals data than the longer TTL [6][14]. Cached is expire, send if modified message 2 nd interest Data Interest If modified message Figure 3.8 Operation of Time To Live Lease Lease cache consistency is illustrated in fig 3.9 (normal case) and 3.10 (updating case). In fig, left side of the fig is the normal condition that has no updating happening in the main server. In that case, the operation procedure is normal as described as follow: 23

32 i) Normal case ( no updating in main server) Step 1: When the consumer request the content, interest will be forwarded until the content is met that may be cached content or content from main server. Step 2: Step 3: And the source will send back the content that the consumer requested. If another consumers request other content or same content, the producer will take place as step 1 and 2. List of cached router Cached Data Interest Cached Figure 3.9 Normal Case of Lease ii) Step 4: Cached content is expire When cached content is replaced by replacement policy and the consumer requests it again, interest will be forwarded till the main server and main server will provide the content. Cache also will be happened accordingly to the cache policies. 24

33 iii) Updating occur in main server Server has dealay for updating Inviladation message to expire the cache 1. Try to update content Cached This router can t reply ack about expire Cached Inviladation message to expire the cache Acknowledge -expire Inviladation message to expire the cache Inviladation message to expire the cache List of cached router Cached Figure 3.10 Updating and delay happened in Lease Step 5: Before content is updated in the main server, all of the cache need to be expire to be cache consistency and to grantee the consumer not to get the stale data. This is confirmed by message between server and the cached router. Server maintain the list of cached router and when it has to update the content, server sends the message to expire the cache to all routers whose cached is still valid. And also all of the routers have to reply acknowledgement to the server that their cached are expire. If one of the routers can t reply acknowledgement to the server, server has to wait till this cache is expired. This makes the delay for updating. If the caching 25

34 time is long, delay for updating is will be long. Although the lease is strong in cache consistency, it has disadvantages in delay time [6][7]. Lease cache consistency has strong cache consistency but it have delay time for content updating that may be long depending on the lifetime of cached. This delay time can be significant disadvantages when the updating of content is important like the stock share, share market and also for disaster application. Consumer also can get the stale data from the cached which can t reply acknowledge to the server. Besides, server has to send the message to invalidate the cached to all the routers which have cached content. Therefore, server has to maintain the list of all routers which has cached content. It will use the server s memory depending on the number of routers. Numbers of routers are huge; the list also will take more memory space. And also sever need updating to update this list. Cached routers also have to reply acknowledge about they expire the cached content. So there are two messages between one router and server before updating and more routers make more messages between them. These messages can make not only network traffic load and bandwidth requirements. Therefore, two significant disadvantages will be in Lease cache consistency, delay time and overhead message between cached router and server [6][7]. 3.3 Proposed Scheme In this research, we approach one kind of cache consistency that base on the Lease cache consistency by overcoming its disadvantages that will explained in this section General Operation of Proposed Lease Scheme In the proposed scheme, some features are added to overcome the disadvantages of Lease cache consistency which are to reduce the overhead message 26

35 between the cache routers and server, to reduce the delay time in updating, to reduce the server s memory required for store list of cached router. Data exptime Data + exptime Send in tag CS table exptime Data exptime exptime Data exptime CS table exptime exptime Figure 3.11 General operation of Proposed Lease Scheme Figure 3.11 shows about the proposed Lease Scheme. The differences between the original Lease scheme and the proposed one are the exptime and the update time. Exptime means that expire time of cache data that will be provided by the Producer when it provide the content. So expire time will be attached with data by tag. When the data with expire time tag is forwarded accordingly to the normal operation to the consumer, one or more routers along the path will cache the copy of data and expire time in the tag. If another interest comes to this cached router, router will provide cached content with the exptime in tag again. The second router will also save both for future use. The main object of this proposed lease scheme is to 27

36 overcome the weakness of original lease. By using expire time provided by the server, server no need to save the list of cache router. In this way, server s memory requirement can be reduced. And sever no need to send the invalidation message to all cached router to expire the cache data and also routers no need to send back acknowledgement about the expire because in proposed scheme, we control the update time in in producer with the expire time of the content in two ways. One is that the update time in the producer is equal to expire time of the cache. It means while the caches expire, updating in the server also happen in same time. Another one is update time is a little larger than expire time that. When the updating happens in the producer, all the caches already expired. So the producer no needs to wait acknowledge and doesn t have to wait all the cache are expired. That s why overhead messages between the producer and cache and the network traffic load can be reduced Calculation of Expire Time in Producer Expire time is the Lease Time of cached content. In the 1 st time of interest request for content, there is no cache in all routers. So interest will be forwarded to the main server. Server will also provide content and expire time attached in packet tag. So server have to calculate expire time accordingly to this equation. (3.1) is expire time of the content value will be in ns and it means when the content will be expired. means that the lifetime of the content that is the duration of how long the content will be stay in cache. means that the time content is provided by the main server. By this equation, we can get expire time of the contents. 28

37 In here, we can make classification to the content accordingly to the application used. There may be much content type like videos, web pages, photo and so on. Depending on the type of contents, we can vary the lifetime of these contents depending on the requirement that is described in section So the expire times are also will be variable accordingly to the lifetime of the contents. Interest Data? Content Type Data Calculation of Exptime Attached in Tag Content with exptime value Figure 3.12 Procedures that the server operates to provide content with expire time Updating in the Producer In the traditional Lease scheme, before the producer, main server updates the content, it needs to send the message to invalid the cached and also the caches have to reply acknowledge messages. In the proposed Lease scheme, we introduce one mechanism to skip this step to eliminate of sending message between the producer and the cached routers. That is to adjust the updating time of the content in server 29

38 with the lifetime of the content. In this research, we only assign one lifetime for all content for a primary stage of simulation, but to be efficient, we can assign the lifetime of content depending on the requirement by adjusting and by classified by type of content. For the optimal solution, Expire time of the content is equal to the Updating time of the content in the server. In this way, when the server need to update, server no need to check the cache is expire or not because of the equation 3.2, all of the cached will be expired in that time. So this is the optimal condition by effectively used of cache. (3.2) Another solution is Expire time is a little smaller than the Updating time as described in equation 3.3. How much smaller is can be varying depending on the requirements. In this part, the value of also can be adjusted by the number of contents in network and the cache size of network. (3.3) Before the server update, all of the caches in routers are already expire accordingly to this equation 3.3. should be as smaller as possible to be the effectively used of cache space. If not, cache is expired so early than the updating time and if there is no content to store in cache, this cache space will be empty. This is not the critical case but it is a kind of wasting the cache space. But, if there is another content cache before the server update, it is ok for using of cache space. With these two equations (3.2) and (3.3), we can control the cached and server update to grantee the consumer not to accept the stale data by expire the cache with the server updating time. And also can eliminate the overhead message between the server and the cached for updating. 30

39 3.3.4 Replacement Policy for Proposed Lease scheme Replacement policy used in proposed scheme is Least Recently Used (LRU) policies, but in here, we change a little to the LRU polices. Traditional LRU polices is to replace the least recently used cached for the next content. But In here, when the new content has to be cached, cache replacement policies will be discard the one whose expire time is expire. To know which one is expired or not, we can check with expire time of the content that also save in content store of the router got from the tag that provided by the producer. In here, we have two solutions. Optimal solution for the replacement policy is : (3.4) Expire Time is means Lease Time of the cached. When expire Time of the cached is expire, this is discard from the CS for save the cache space So that will be the optimal solution for effectively used of cached space. Another solution is : (3.5) As in equation 3.3, also should be small as much as possible. If it is large, replacement policy will replace with the other content. In that case, if another request will come for this content, interest will have to main server and server will provide the content. If this situation happened frequently, the network traffic load will be large. This will be effect to the main purpose of caching. So we make the Replacement Time is less than the Expire Time. If the Replacement Time is greater than the Expire Time, the cached is already expired but not removes from CS table, and there is no replacement. So the consumers 31

40 have a chance to get the stale data from the cache. That s why, in equation 3.5, replacement time is smaller than expire time of the cache. Chapter 4 CACHE CONSISTENCY: LEASE 4.1. Introduction for simulation We present here about the simulation for proposed Lease. First step is about in producer, main serve additional function for this proposed one. And second is about the tag and third one is for forwarder. And then how to do the replacement to save the cache size effectively that is also the key point of this proposed Lease Simulation Challenges in Producer to provide content with tag Content Specification in Producer As mention in section 3.3.2, producer has to provide, expire time for the data. In here, we will describe how this is provide expire time for the content. Following table is the example of how the content s lifetime is assigned and this have to be predefined by the content owner or publisher. Table 4.1 Content Name with Lifetime Content Name Type Lifetime (s) /parc.com/videos/widget.mpg/_v<timestamp>/_s1 video 10 /upc.edu/fcoras/presentation.pdf/_v<timestamp>/_s2 Pdf 5 /waseda.jp/gits/naka_lab/m2/ccn.ppt/_ver7/_s2 Power point 7 32

41 Table 4.1 is just the example of how the lifetime is assign to the content name for evaluate in Producers. When the interest is received by producer in the condition of no cache in all routers in first stage, producer has to send back the data in reversed path to the consumer. In normal traditional Lease, producer only provide the data to the next step. But in our proposed lease, the producer has to evaluate the lifetime of the content to provide expire time (Lease Time) of the content. Interest Producer Data? Content Type Data Calculation of Exptime Data Packet Create Tag Set exptime to tag Net-Device-Face Content with exptime value Figure 4.1 Producer and the net device face how to operate for proposed Lease 33

42 4.3. Simulation Challenges with Content Store. On Receiving Data Data Copy With Data Data Expire Time value CS Figure 4.2 To store expire time in CS, on Receiving Data CS Data Exptime Set exptime in Tag Data Packet with exptime Tag Figure 4.3 To get expire time in CS, on Outgoing Data 34

43 Content store also has to save expire time when the data is insert to invalid the content. If there is another request come to it, cache is already. So, it will provide the cache and expire time for the next cache. Next cache router also has to save this expire time value for the cached content. This is shown in fig 4.2 and 4.3 In this way, all of the caches can hold expire time for each contents respectively and can be invalid at the same time for each contents. 4.4 Simulation Challenges of Discard entry and Replace the cache Additional part for our proposed lease is that to discard the content whose expire time is expire from the CS. This is described in figure 4.4 according to equation (3.4). Start Check CS Exp time is expire? No Yes Discard this Entry Figure 4.4 Producers to discard the CS entry whose content is expire 35

44 In figure 4.4, we check expire time of the content, if it is expire, discard it from the CS. By this way, we can get the place to save the new content. That makes the cache effectively used. Another additional solution is for replacement when the new content is need to cache in the CS and CS is full. In that case, one of the caches will be replaced to save the new content. It is to search the smallest expire time value in whole CS entry in first. The smallest expire time is means that one will be expire soon than the others. And replace with the new content that has to cache. By this way cache can be used effectively. Start Check entire CS (the smallest one in expire time) Smallest expire time from CS Replace this Entry Figure 4.5 Producers to replace the CS entry 36

45 4.5 Algorithm for Overall Operation Algorithm Producer : For each interest Set Lifetime according to Content Types Check data Send data Calculate Exptime Net-Device-Face Make packet for data Attach Tag to data packet Call function -calculate Exptime- Set Exptime to tag Content Store: For each entry Create CS table (Exptime) Forwarder: On Receiving Data Data Copy with Packet Entry to CS table (Exptime) :On Outgoing Data Set Exptime to tag Send data packet with tag Replacement: For each Replacement While (whole Cs entry) do If Exptime=simulator now ( ) THEN Discard this entry 37