An Efficient Link Bundling Transport Layer Protocol for Achieving Higher Data Rate and Availability

Transcription

1 An Efficient Link Bundling Transport Layer Protocol for Achieving Higher Data Rate and Availability Journal: IET Communications Manuscript ID: COM Manuscript Type: Research Paper Date Submitted by the Author: 10-May-2011 Complete List of Authors: Hashmi, Muhammad; University, Computing & Technology Rehman, Raza; Iqra, Computing & Technology Shafi, Imran; Iqra, Computing & Technology Shah, Ismail; Iqra, Computing & Technology Ahmad, Jamil; Iqra, Computing & Technology Keyword: NETWORK MANAGEMENT, NETWORK PROTOCOLS, NETWORK SERVERS, NETWORK TRAFFIC

2 Page 1 of 16 IET Communications An Efficient Link Bundling Transport Layer Protocol for Achieving Higher Data Rate and Availability Muhammad Adnan Hashmi, Raza-ur-Rehman, Imran Shafi, Syed Ismail Shah, Jamil Ahmad Department of Computing and Technology Iqra University, Islamabad Campus, Pakistan hashmi_adnan@yahoo.com, [raza, imranshafi, ismail, jamil]@iqraisb.edu.pk Abstract: By using aggregation of more than one physical interface enables aggregated links to work at the same time for enhancementof bandwidth and data speed beyond capability of single link with additional feature of resilience. However, inappropriate schemes may cause undesirable network behavior and thus requires high efficiency and reliability in its implementation. A new link bundling transport layer protocol is presented which ensures efficient and reliable trunking of multiple wired or wireless network interfaces resulting in escalation of data rate. The proposed scheme is found efficient, reliable and independent of available paths [8] by implementing the Luby transform (LT) codes. The LT codes do not require any interface characteristics management prior to transfer of data which helps in aggregation of interfaces having different characteristics in less time. Also, the data flows seamlessly even if one of the link experiences fluctuation or complete shutdown. The proposed protocol requires less processing requirement, minimum connection establishment time and no interface management as compare to any TCP based protocol. Experimental results demonstrate the effectiveness of the approach. 1. Introduction Transport layer protocol use only one interface or connection between sender and receiver for data transfer. If more than one paths need to be established using multiple interfaces between sender and receiver then there is a requirement of efficient management of links and their aggregation for creation of single logical link. Aggregation of links demands the reliability and

3 Page 2 of 16 efficiency i.e. data must be transferred without loss with minimum time and less overhead. Efficiency is an important factor especially in case of erasure channels e.g. wireless. We find few approaches in the literature to address these issues partially. One of the popular approaches is the parallel TCP (ptcp) [1]. It is the extension of TCP which utilizes the concept of virtual TCP (TCP-v) and Strip Manager (SM) and increases the bandwidth by aggregating the multiple paths. TCP-v deals with the per path functionality within single logical link and SM manage all TCP-v for successful transfer of data. On the other hand transaction TCP (T/TCP) [2] skips 3-way handshake process by introducing connection count state which ultimately helps in reducing time factor required for connection establishment. Skipping 3-way handshake is also known as TCP Accelerated Open (TAO). This helps in saving overall time required for transmission of data but has some security threats which are addressed in enhance transaction TCP (ET/TCP) [2]. Another approach is link aggregation control protocol [5] which increases the bandwidth by aggregating wired interfaces. It has different modes for traffic distribution over multiple interfaces. IEEE 802.3ad standard is related to link aggregation whereas IEEE 802.1AX-2008 represents the redundancy and reliability. The approaches like T/ TCP, ET/TCP and ptcp are transport layer implementations whereas link aggregation is implemented at the data link layer. The objective of escalation ofdata rate can be achieved withsome special features at transport layer like efficient memory management both at sender and receiver side, paths selection and interface management especially in case of short term fluctuation and shutdown at transport layer of custom protocol. Better data rate is the requirement of almost every application connected to server for services either with video streaming server, proxy server, webserver etc. Server with high processing power can t play an important role if the via is not able to support the

4 Page 3 of 16 IET Communications application s data flow. Currently network devices like switches etc. don t have any built in facility installed by vendors for the aggregation of interfaces in attainment of escalation in data rate. Aggregation of network interfaces depends on the characteristics of interfaces i.e. bandwidth, delay and loss especially in case where wired and wireless links are involved. Other than major characteristics, short term fluctuation or even complete shutdown on any of the interface can affect the overall transfer of data. Path independence problem is achieved by implementing the LT codes. The LT codes are rate-less erasure code and are first practical implementation of the Fountain codes [4] [7]. These codes are well suitable especially in case where network has higher latencies and higher loss rates provided multiple interfaces are available. The proposed solution in this paper introduces a custom transport layer protocol developed using C/C++ on Ubuntu platform. It utilizes UDP packets stream and for reliability of connection, three way handshake process is implemented before the establishment of actual data stream. The data packets are passed through LT encoder at sender side whereas LT decoder decodes the same till the retrieval of required data. The paper is divided in five sections. Section 2 discusses the proposed link bundling transport layer protocol. Section 3 presents the implementation details and discussion on experimental results is given in section 4. Finally section 5 concludes the paper. 2. The Proposed Link Bundling Transport Layer Protocol The main issue for achieving aggregate bandwidth is the management of network interfaces. Each network interface has different characteristics like bandwidth, delay and loss. The links aggregation of different characteristics require efficient algorithm for traffic distribution on

5 Page 4 of 16 multiple interfaces. Short term fluctuation can be experienced by any of the link during data transfer. In addition to fluctuation, link can also experience complete shutdown. For successful transfer of data aggregated links are managed in such a way that if fluctuation or shutdown experience by any one of the link, other links remain up without any problem, this situation definitely increase the overall time require for complete transfer of data but provide enhanced reliability and redundancy [9]. Wireless Links Wired Links Sender LT Codes Encoder Links Aggregation LT Codes Decoder Receiver Fig 2.1 Proposed Infrastructure Model Fig 2.1 provides overview of the proposed model. Sender and receiver having multiple wired and wireless interfaces are combined using link aggregation to a single logical link where data is transferring over multiple paths with the implementation of LT codes.

6 Page 5 of 16 IET Communications Figure 2.2 Server Flow Chart

7 Page 6 of 16 Figure 2.3 Client Flow Chart Fountain codes [7] are chosen for the successful transfer of data. These codes are independent from path. It works on the principle that enough should be received to recover exact copy of

8 Page 7 of 16 IET Communications sender data regardless of loss of encoded symbols. Fountain codes require efficient memory management especially on receiver side. LT codes have been selected for successful transfer of data reliably and efficiently. They can generate limitless output symbols from the fixed number of message symbols. LT codes recovery and delivery doesn t depend on network reliability; they have linear and minimum encoding complexity with iterative encoding decoding algorithm. For reliable connection establishment three way handshakes process is established between sender and receiver. Packets used for connection establishment and data transfer are distinguished with the help of its type field. After successful handshake process, information about the transfer data is synchronized between sender and receiver. The next phase is LT encoding process, packets first converted into encoded symbols and then transferred over multiple interfaces. Receiver applies the LT codes decoder operation on the encoded symbol till the exact copy of sender s data is retrieved on its side. Successful transfer of data leads to the termination of connection. 2.1 Implementation The proposed link bundling transport layer protocol was developed using C/C++ and for the implementation purpose Ubuntu platform is chosen. LT codes are implemented using linked list for memory management at both sender and receiver side. C/C++ is the higher level language which utilizes the memory efficiently. Most of the application servers are developed using C/C++. Its performance over UNIX is far better than any of other operating systems. GNU compiler includes the libraries for the C/C++, it is free ware and used in the UNIX distributions. UNIX is one of the best operating system having lots of flavor, Ubuntu is one of them.

9 Page 8 of 16 3-way Handshake (using Control Packet) Exchange Information about Required Data (using Control Packet) Apply LT Encoder (Encode Data Packets) Data Packets Streaming (Data Packet + Control Header) Apply LT Decoder (Decode Data Packets) Terminate Connection (using Control Packet) Fig Protocol Flow Diagram The implementation can be categorized in two major areas, one part is related to the connection establishment that simulates the same handshake behavior used by the TCP whereas second part deals with the actual transfer of data packets on multiple interfaces, it also ensure the data integrity, reliability and efficiency with the use of LT codes.

10 Page 9 of 16 IET Communications 1. Bind Sockets on multiple available interfaces 2. Bind Sockets on multiple available interfaces (Handshake Started) 3. Send Connection Request 4. Receive Connection Request 5. Send Response 6. Receive 7. Send Ack 8. Receive Ack SERVE (Handshake End) 10. Receive File Request 11. Send File Info (Name, Size) 9. Send File Request CLIENT 12. Receive File Info 13. Reading File/Data from storage 14. Apply Luby Code Encoding 15. Sending Data Packets on Multiple Interfaces 16. Receiving Data Packets on Multiple Interfaces 17. Apply Luby Code Decoding 18. Writing File/Data on storage 19. On Completion of Data Close the Connection Figure Detail Data Flow Diagram between Sender/Server and Receiver/Client

11 Page 10 of 16 For successful transfer of data, one will act as server and other as client. Server binds the sockets with all the available interfaces, it will enable clients to initiate connection request. Sockets are basically the file descriptor in Unix/Linux which helps in communication between the two programs. Client sends the request to server for establishment of connection. The unsigned integer is used to distinguish packets i.e. either its request, response or acknowledgment packet. Packet type filed is also reserved which is used for representing control packet or actual data packet. Server receives request from the client andinitiate the handshake process by sending response back to the client. Handshaking is the process by which two devices initiate the communication. This response of the server is again acknowledged by the client by sending the acknowledgment packet and successfully completes the handshake process. Server Client Handshake Started 1. Connection request 2. Response ofrequest 3. Acknowledgment Handshake Completed Figure way Handshake Process After successful handshake, client sends the request for the data e.g. name of the file etc. The request packet contains the type of packet and name of file.in response of the client request server will send complete information of data e.g. file name and size of fileused for the

12 Page 11 of 16 IET Communications calculation of total number of chunks. The data packet size is already decided between the client and server. Requisite data or file first divided into equal size of chunks and applied to the LT codes encoder, the resultant packets will start flowing towards client on multiple available paths. The length of the symbols is chosen as 512 bytes plus the header bytes. After selection of randomly chosen degree d which defines the number of packets going to be XORed for the composition of encoded symbol, header is also attached with the encoded symbol having information of distinct chosen neighbor indices selected with the help of random generator and chosen degree d. Client who receives these encoding symbols start applying the LT codes decoding process. The receiver or client with the help of header get degree d and number of distinct indices and start the decoding process. Encoded symbol with degree one is the actual copy of sender data. Any Encoded symbol XORed with already decoded symbol reduces the degree d of encoded symbol by one. Same process repeats until exact copy of sender s data is decoded at the client side. After successful retrieval of decoded symbols at the client side, connection isterminated by both ends. 3. Results and Discussion In this section we present the results showing the performance of an efficient link bundling transport layer protocol using two hosts connected with two separate links aggregated in form of single logical link. 1 Gbps Host 1 Gbps Host Figure 3.1 Aggregated links between hosts

13 Page 12 of 16 We use two hosts directly connected with each other as shown in figure 3.1, they use two wired interfaces both having speed of 1Gbps.Hosts utilizes both the interfaces at the same time for successful transfer of data using the protocol implemented at each side. Figure 3.2 Escalation of Bandwidth Figure 3.2 shows that same amount of data take less time in case of two interfaces as compare to single interface. The result enforces our argument that using efficient aggregation of multiple interfaces can achieve escalation in bandwidth.

14 Page 13 of 16 IET Communications Esclation of Data Rate Time (in mliseconds) Interfaces - 01 Interfaces Number of Packets Figure 3.3 Comparison of Data Rate againsttime Figure 3.3 shows the variable increase in bandwidth when size of data grows for the same aggregated link as compare to data with lesser packets, that means better performance for large files. Figure3.4 Overhead Packets

15 Page 14 of 16 Figure 3.4 shows that while transferring of data there is always acceptable overhead of encoded symbol. Data consisting of more packets have more overhead as compare to data having lesser packets. Overhead Received Overhead Encoded Packets at Receiver Encoded Packets Overhead Figure 3.5 Overhead Received against Encoded Packets Figure 3.5 shows the overhead packets received against the total number of actual data packets. This approach can play an effective role especially in case of video streamed application where bandwidth or data rate does matter as come to overhead. Faster the frames received at the receiver side better will be the quality and experience of user. It is clear from the simulation results that increase in data rate produces better results in terms of time it will take for transfer of data using two interfaces but it is also showing that overhead is growing exponential that can be reduced with the implementation of online codes [6].

16 Page 15 of 16 IET Communications 4. Conclusion We have presented a new link bundling transport layer protocol which ensures increase in bandwidth along with higher availability with efficiency and reliability. This approach can also be used to increase uplink traffic between any two switches where more than one interfaces are available. It helps in increasing the bandwidth without upgrading the equipment. Experimental results demonstrate the effectiveness of approach. In future, the LT codes may be embedded in Linux kernel as it is expected to exhibit optimized and better results. Also, the cost of overhead may be reduced by introducing the online codes instead of the LT codes. References [1] H.Y. Hsieh and R. Sivakumar, ptcp: An End-to-End Transport layer Protocol for Striped Connection, in Proc.10 th IEEE International Conference Network Protocols, 2002, pp [2] R. Bin and Z. Xialoan, Enhanced Transaction TCP Design and Implementation, Internet: [Apr 22, 2011] [3] David D. Clark, The Design Philosophy of the DARPA Internet Protocols, SIGCOMM Computer Communication Review, vol. 18, no. 4, pp , Aug 1988 [4] Michael Luby, LT Codes, in Proc. 43 rd Annual IEEE Symposium on Foundation of Computer Science, 2002, pp [5] Syskonet, Link Aggregation according to IEEE 802.3ad., Internet: [May. 3, 2011]

17 Page 16 of 16 [6] P. Maymounkov, Online codes", NYU Technical Report TR , Nov 2002 [7] D.J.C. MacKay, Fountain codes, in Proc. IEEE Proceedings in Communication, vol. 152, no. 6, 2005, pp [8] J.W. Byers, M. Luby, M. Mitzenmacher, A. Rege, A Digital Fountain Approach to Reliable Distribution of Bulk Data, in Proc. ACM SIGCOMM, 1998, pp [9] Linux Foundation, bonding., Internet: [Feb 25, 2011]