Router-based Content-aware Data Redirection for Future CDN Systems

Published Online June 2014 in MECS DOI: 10.5815/ijcnis.2014.07.01

Communication technologies have become significantly advanced in the past few decades, with new technologies being invented to achieve efficient communication. In addition, people are more interested in sharing knowledge and information for various purposes. Innovations in information sharing are continuously accelerated to cater user needs in such environments. These motives have encouraged the construction of sophisticated environments for effective communication and information delivery.

Content Delivery Networks (CDNs) are implemented to achieve low-latency content delivery such as; data streaming, on-line gaming and e-commerce web accesses by placing content servers near the customer [1]. RR techniques are used in CDNs to redirect client requests to the nearest surrogate server [1-7]. Therefore, it is important to maintain a better RR method that can find the nearest server for a particular user. Besides, a RR method can also achieve efficient server load balancing [6, 10]. DNS redirection and URL rewriting [1] are the most commonly used CDN redirection methods in the current CDNs.

Several studies have been conducted on the impact of content delivery on networks in the past few decades. Before the adoption of worldwide server caching, such as CDN, some early research on the effectiveness of interior web caching were performed by Gadde et al [8]. In addition, recent studies [9, 10] indicate the DNS-based method can significantly reduce the download latency. Top-level CDN providers, such as Akamai [3, 4] and CenturyLink [5], use DNS-based RR to redirect client requests to the nearest surrogate server, or sometimes known as a redirector or an edge server. However, all those studies use multi tire name servers in order to maintain the consistency of changes of surrogate servers [3, 6, 11].

The DNS-based redirection use small Time to Live (TTL) values at the clients in order to adapt the changes occurs on the CDN infrastructure [1, 3, 4]. Consequently, small TTL values lead the client to access the local name server frequently to get an up to date infrastructural changes [1, 11 - 13]. Furthermore, name server resolving process increases the client waiting time due to communication between local name server and authoritative name servers to find the CDN surrogate servers’ IP addresses. Poese et al. [6] proposed a content aware traffic engineering design that give some hints about the ISP collaborated method of redirecting packets based on both status of the servers and the network. But that study also was not able to provide a solution to the clients regarding the latency factors of connection initiation and lags of the connection due to the DNS resolving process and small TTL values. Therefore, we argue that the usage of multi tire name servers and small Time to Live (TTL) values on DNS-based RR considerably degrades the connection initiation process.

We argue that, if an edge router of the network can identify the server locations and redirects client request to the servers regardless of the IP address of the final destination, we can utilize data redirection process to achieve faster connection initiation and faster adaptation to the server changes in a CDN. To this end, this study proposes a redirection mechanism which is redirecting

Fig. 1. SoR-based RR Overview packets based on the content of the packet by using an Service-oriented Router [14, 15, 16] as an edge router of a network. Furthermore, we confer that the basic functions of the SoR such as; DPI and content based data redirection, introduce the ideal test bed to implement a prototype for an edge router based date redirection. Fig. 1 illustrates the basic concept of SoR-based RR. The client sends the data stream to the SoR which is placed at the edge of network and SoR selects the proper surrogate; A, B, C or D, based on the content of the packet and the status of the server and, redirects the packet stream to the selected server. In addition, an SoR redirects packets by performing server load balancing if an SoR has routing records of more than one surrogate server.

The paper is arranged as follow. A review of the DNSbased RR is provided in the Section 2. Section 3 describes the motivation of this study. Section 4 describes the software implementation of the SoR followed by the detailed explanation of prototype implementation of the SoR-based RR in Section 5. Test cases designs, implementations and the test results are presented in the Section 6. Section 7 is used to discuss the effectiveness of the SoR for the CDN redirection process and how the SoR-based RR is able to overcome the limitations of existing methods. Section 8 concludes the paper with tentative future implementations.

• II.    background study

The Domain Name System (DNS) is a distributed database of records (name-to-address mapping) cached at intermediate name servers [17]. Each record has a time-to-live (TTL) value that indicates how long it will be cached. Generally TTL values are set to one day [17]. The original operation of the DNS is to resolve a Fully Qualified Domain Name (FQDN) to its corresponding IP address [11, 17].

However, in CDN infrastructures, once the DNS is used to resolve the nearest surrogate server’s IP for a particular end-user, the DNS cache becomes a significant obstacle to providing real-time control of the CDN. The name resolution cache must be disabled by setting TTL values to zero in order to obtain real-time control in

CDNs [1]. Small TTL values allow fine-grained load balancing and rapid response to changes in the server or network load. Nonetheless, small TTL values force clients to contact the authoritative name server for every name resolution request. However, this phenomenon increases access latency for particular content.

To understand the behavior of the DNS in a real network, we captured and analyzed the traffic of a major Internet backbone network for academics in Japan. The most noticeable fact was that the occupation ratio of the DNS, which was occupied more than 25% of traffic during the captured time. Though we cannot conclude that the measured DNS traffic occupation was entirely from CDN networks, this clearly highlights that the DNS traffic is high in the Internet data exchange.

On the other hand, recent studies [6, 18, 19] have been done to research about the ISP involvement of the DNS redirection. Authors have theorized that the ISP-CDN and DNS collaboration make the ideal network infrastructure for faster data delivery on the Internet. Those studies suggested of placing low level name servers at the ISP. Then redirect clients to the nearest server based on the status of both ISP’s network the status of the content servers. However, this method also requires the clients to frequently contact local name server in order to get the status of the CDN upon the expiration of TTL values [1, 13, 19]. Besides, Anees et al. [1] confirmed that the DNSbased server redirection must be based on the name server’s location, not on the client’s location. Therefore, the response delay associated with connection initiations and request redirections depends on the number of hops to the authoritative name server. That study pointed out that aforementioned factors are significant factors in data redirection and client response time.

• III.    motivation