Globule: A collaborative Globule: A collaborative Content Delivery - PDF document

Globule: A collaborative Globule: A collaborative Content Delivery Network Content Delivery Network Guillaume Pierre Maarten van Steen Presented By: Marwa K. Elteir Outline Outline Introduction Introduction Related work Related work Collaborative Content Delivery Network Collaborative Content Delivery Network � Requirements Requirements � � Model Model � � Content Distribution Content Distribution � � Management Management � Handling dynamic content Handling dynamic content Comments Comments 1

Introduction Introduction For a user or an organization to be able to For a user or an organization to be able to publish a web site , they either do that , they either do that publish a web site through paying a monthly fees to a hosting through paying a monthly fees to a hosting servers or deploying their own servers. servers or deploying their own servers. Problems appear when the web sites Problems appear when the web sites owners requires some sort of guarantees owners requires some sort of guarantees about the availability about the availability and and performance performance of of their web sites. their web sites. Introduction Introduction One solution is to use a commercial commercial One solution is to use a Content Delivery Network (CDN) which (CDN) which Content Delivery Network provides resources to be allocated provides resources to be allocated dynamically to serve certain web sites. dynamically to serve certain web sites. Another solution is to build a personal personal Another solution is to build a CDN composed of the resources available CDN composed of the resources available locally in the user's side. locally in the user's side. 2

Introduction Introduction The work of this paper extends the idea of The work of this paper extends the idea of personal CDN into what is called personal CDN into what is called Collaborative CDN (CCDN). Collaborative CDN (CCDN). In CCDN, an overlay network of end- In CCDN, an overlay network of end -user user machines cooperate through a wide area machines cooperate through a wide area network in a peer- -to to- -peer fashion peer fashion. . network in a peer They provide some of their resources and They provide some of their resources and in return be able to replicate their web in return be able to replicate their web content across the CCDN which guarantee content across the CCDN which guarantee the required performance. the required performance. Related Work Related Work Kazaa and BitTorrent Coral- CoDeeN � Replicas � Reasonable performance � No efficient replica placement scheme � No efficient handle for dynamic content ACDN � Dynamic content � Assume no data modification DotSlash � Dynamic content � Assume the database is not bottleneck 3

CCDN Requirements CCDN Requirements Transparency to the clients. Transparency to the clients. Automatic redirection to the target replica. Automatic redirection to the target replica. Network proximity between clients and Network proximity between clients and their replicas should by used as a factor to their replicas should by used as a factor to distribute the replicas. distribute the replicas. Consistency management. Consistency management. CCDN Model CCDN Model 4

CCDN Content Distribution(1/6) CCDN Content Distribution(1/6) Measuring proximity Measuring proximity The locations of replicas are determined The locations of replicas are determined based on their proximity to the clients. based on their proximity to the clients. They chose internode internode latency latency as their as their They chose proximity measure. proximity measure. This measure is also used to redirect the This measure is also used to redirect the client to the most appropriate replica. client to the most appropriate replica. CCDN Content Distribution(2/6) CCDN Content Distribution(2/6) Measuring proximity Measuring proximity Estimating the distance between every Estimating the distance between every client machine to the replica servers is client machine to the replica servers is impossible. impossible. An alternative is to assume that the nodes An alternative is to assume that the nodes are placed in M- -dimensional geometry dimensional geometry are placed in M space. (M =6) space. (M =6) The coordinates of a node X is calculated The coordinates of a node X is calculated based on the latency between this node based on the latency between this node and m landmark landmark nodes nodes and m 5

CCDN Content Distribution(3/6) CCDN Content Distribution(3/6) Measuring proximity Measuring proximity 90 % of the latency estimation fall in the interval [2/3 L, 3/2L], where L is the actual latency between two nodes. CCDN Content Distribution(4/6) CCDN Content Distribution(4/6) Replica Placement Replica Placement The clients are localized. (Dynamic process?) The clients are localized. (Dynamic process?) The M- -dimensional space is partitioned into cells dimensional space is partitioned into cells The M of equal size. of equal size. Each cell is ranked according to the number of Each cell is ranked according to the number of clients located in this cell. clients located in this cell. Put the replica servers in the k highest ranked Put the replica servers in the k highest ranked cells. cells. The results of this algorithm is near optimal. In The results of this algorithm is near optimal. In addition, its computational complexity is lower addition, its computational complexity is lower than the optimal alternative. than the optimal alternative. 6

CCDN Content Distribution(5/6) CCDN Content Distribution(5/6) Client Redirection Client Redirection HTTP based redirection HTTP based redirection � Per Per- -page basis decision page basis decision � � No transparency .. URL is returned No transparency .. URL is returned � DNS based redirection DNS based redirection � Per Per- -site basis decision site basis decision � � Transparent .. IP is returned Transparent .. IP is returned � Another redirection policy can be used other Another redirection policy can be used other than the estimated latency like for example the than the estimated latency like for example the average load in each replica server. average load in each replica server. CCDN Content Distribution(6/6) CCDN Content Distribution(6/6) Content availability Content availability Redirector availability : DNS protocol allows : DNS protocol allows Redirector availability multiple redundant servers to be registered for multiple redundant servers to be registered for the same name. the same name. Replica server availability : Redirector server : Redirector server Replica server availability should periodically check the availability of should periodically check the availability of replica servers. replica servers. Fresh copy availability : Every replica server : Every replica server Fresh copy availability should be able to contact the origin server or the should be able to contact the origin server or the backup server to retrieve the up- -to to- -date date backup server to retrieve the up documents. documents. 7

CCDN Management CCDN Management Security Security � Malicious content .. Sandboxing technique Malicious content .. Sandboxing technique � � Malicious replica server .. Client solution or Malicious replica server .. Client solution or � Origin server solution Origin server solution Brokerage Brokerage � Allows any server to perform automatic Allows any server to perform automatic � resource discovery and allocation. resource discovery and allocation. � Important especially during the registration Important especially during the registration � process. process. Supporting Web Applications (1/2) Supporting Web Applications (1/2) 8

Supporting Web Applications (2/2) Supporting Web Applications (2/2) Comments Comments The main idea is based on deploying a peer- -to to- - The main idea is based on deploying a peer peer overlay network. It is adopted a lot as a peer overlay network. It is adopted a lot as a solution to several problems in distributed solution to several problems in distributed systems arena. systems arena. How they handle the dynamic nature of the web How they handle the dynamic nature of the web clients and replicas? clients and replicas? How the assignments of the different servers to How the assignments of the different servers to the participating nodes are achieved? the participating nodes are achieved? Performance evaluation Performance evaluation 9

Thank you Thank you 10