• Proceedings of the Korean Information Science Society Conference
• /
• 2003.10c
• /
• pp.106-108
• /
• 2003

기존 TCP 기술은 높은 대역폭(High­Bandwidth) 및 큰 전송지연(High Delay)을 가진 통신에는 적합하지 못하다. TCP 기술의 성능향상을 위한 방법으로 TCP 제어 알고리즘을 수정하는 방법과 TCP Tuning 방법이 있다. 본 논문에서는 TCP Buffer Tuning 기술에 초점을 맞춰 통신망 상황에 따른 응용프로그램별로 자동화된 Buffer Tuning 기법을 제공하는 기술을 제안한다. ATBT(Automatic TCP Buffer Tuning) 에서는 송신측의 Buffer 크기를 조절하여 성능향상을 나타냈고, DRS(Dynamic Right Sizing)에서는 수신측의 Buffer 크기를 조절하여 성능향상을 도모하였다. 본 논문에서는 ATBT와 DRS의 장점을 접목하여 구현함으로써 보다 나은 성능향상을 나타내고 각 송.수신측의 모든 연결에 대해서는 Buffer를 공평하게 할당하여 메모리 사용의 효율을 높이고자 한다.

• The KIPS Transactions:PartC
• /
• v.12C no.1 s.97
• /
• pp.121-128
• /
• 2005

Tho existing TCP(Transmission Control Protocol) is known to be unsuitable for a network with the characteristics of high RDP(Bandwidth-Delay Product) because of the fixed small or large buffer size at the TCP sender and receiver. Thus, some trial cases of adjusting the buffer sizes automatically with respect to network condition have been proposed to improve the end-to-end TCP throughput. ATBT(Automatic TCP fluffer Tuning) attempts to assure the buffer size of TCP sender according to its current congestion window size but the ATBT assumes that the buffer size of TCP receiver is maximum value that operating system defines. In DRS(Dynamic Right Sizing), by estimating the TCP arrival data of two times the amount TCP data received previously, the TCP receiver simply reserves the buffer size for the next arrival, accordingly. However, we do not need to reserve exactly two times of buffer size because of the possibility of TCP segment loss. We propose an efficient TCP buffer tuning technique(called TBT-PLR: TCP buffer tuning algorithm based on packet loss ratio) since we adopt the ATBT mechanism and the TBT-PLR mechanism for the TCP sender and the TCP receiver, respectively. For the purpose of testing the actual TCP performance, we implemented our TBT-PLR by modifying the linux kernel version 2.4.18 and evaluated the TCP performance by comparing TBT-PLR with the TCP schemes of the fixed buffer size. As a result, more balanced usage among TCP connections was obtained.

• Journal of Internet Computing and Services
• /
• v.8 no.1
• /
• pp.15-23
• /
• 2007

Wireless LAN (IEEE 802.11) uses traditional TCP for reliable data transmission, But it brings the unintentional packet loss which is not congestion loss caused by handoff, interference, and fading in wireless LAN. In wireless LAN, TCP experiences performance degradation because it consumes that the cause of packet loss is congestion, and it decrease the sending rate by activating congestion control algorithm. This paper analyzes that correlation of throughput and buffer size for wireless buffer tuning. We find MBT (Maximum Buffer Threshold) which does not increase the throughput through the analysis, For calculation of MBT, we experiment the throughput by using high volume music data which is creased by real-time performance of piano. The experiment results is shown that buffer tuing based on MBT shows 20.3%, 21.4%, and 45.4% throughput improvement under 5ms RTT, 10ms RTT, and 20ms RTT, respectively, comparing with the throughput of operation system default buffer size, In addition, we describe that The setting of TCP buffer size by exceeding MBT does not have an effect on the performance of TCP.

• Journal of Internet Computing and Services
• /
• v.10 no.4
• /
• pp.199-211
• /
• 2009

Recent advancements in wireless networks have enabled support for mobile applications to transfer data over heterogeneous wireless paths in parallel using data striping technique [2]. Traditionally, high performance data transfer requires tuning of multiple TCP sockets, at sender's end, based on bandwidth delay product (BDP). Moreover, traditional techniques like Automatic TCP Buffer Tuning (ATBT), which balance memory and fulfill network demand, is designed for wired infrastructure assuming single flow on a single socket. Hence, in this paper we propose a buffer tuning technique at senders end designed to ensure high performance data transfer by striping data at transport layer across heterogeneous wireless paths. Our mechanism has the capability to become a resource management system for transport layer connections running on multi-homed mobile host supporting features for wireless link i.e. mobility, bandwidth fluctuations, link level losses. We show that our proposed mechanism performs better than ATBT, in efficiently utilizing memory and achieving aggregate throughput.

• Journal of KIISE:Computer Systems and Theory
• /
• v.33 no.7
• /
• pp.363-369
• /
• 2006

It is aware that Bandwidth management with dramatically increasing traffic on account of frequent and large file transmission in a data grid environment is one of essential needs. A this paper we propose new method which guarantees QoS (Quality of Service) by being in control of resources in TCP layer based on existing studies that manage bandwidth over TCP buffer tuning. General QoS solutions manage network resources subsequent to observing them in IP or link layer, but the scheme in the paper is able to control network resources in TCP layer that is network upper layer. Consequently, bandwidth allocation to each user can be efficiently controlled depending on an authority each user is given so that users could be use different bandwidth. It is expected that a new paradigm is supposed in network resource management and the method of levies for users' bandwidth uses.

• Journal of Internet Computing and Services
• /
• v.7 no.1
• /
• pp.131-141
• /
• 2006

Recently, many researchers have been studied several high performance data transmission techniques such as TCP buffer Tuning, XCP and Parallel Sockets. The Parallel Sockets is an application level library for parallel data transfer, while TCP tuning, XCP and DRS are developed on kernel level. However, parallel socket is not analyzed in detail yet and need more enhancements, In this paper, we verify performance of parallel transfer technique through several experiments and analyze character of traffic interference among socket connections. In order to enhance parallel transfer management mechanism, we predict the number of socket connections to obtain SLA of the network resource and at the same time, affected network bandwidth of existing connections is measured mathematically due to the interference of other parallel transmission. Our analytical scheme predicts very well network bandwidth for applications using the parallel socket only with 8% error.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.12 no.11
• /
• pp.5179-5202
• /
• 2018

Transmission Control Protocol (TCP) is the most widely used protocol in the cloud data centers today. However, cloud data centers using TCP experience many issues as TCP was designed based on the assumption that it would primarily be used in Wide Area Networks (WANs). One of the major issues with TCP in the cloud data centers is the Incast issue. This issue arises because of the many-to-one communication pattern that commonly exists in the modern cloud data centers. In many-to-one communication pattern, multiple senders simultaneously send data to a single receiver. This causes packet loss at the switch buffer which results in TCP throughput collapse that leads to high Flow Completion Time (FCT). Recently, Software-Defined Networking (SDN) has been used by many researchers to mitigate the Incast issue. In this paper, a detailed survey of various SDN based solutions to the Incast issue is carried out. In this survey, various SDN based solutions are classified into four categories i.e. TCP Receive Window based solutions, Tuning TCP Parameters based solutions, Quick Recovery based solutions and Application Layer based solutions. All the solutions are critically evaluated in terms of their principles, advantages, and shortcomings. Another important feature of this survey is to compare various SDN based solutions with respect to different performance metrics e.g. maximum number of concurrent senders supported, calculation of delay at the controller etc. These performance metrics are important for deployment of any SDN based solution in modern cloud data centers. In addition, future research directions are also discussed in this survey that can be explored to design and develop better SDN based solutions to the Incast issue.

• The Journal of Korean Association of Computer Education
• /
• v.12 no.2
• /
• pp.77-86
• /
• 2009

Large-scale applications, that needs large-capacity R&D resources and realtime data transmission, have demanded more stable and high-performance network environment than current Internet environments. Recently, global R&D networks have focuses on utilizing Lambda networking technologies and resource reservation systems to be satisfied with various applications' requirements. In this paper, we modify the existing DRS (Dynamic Right-Sizing) model to reflect various advantages in terms of the stability and high-capacity of Lambda network. In addition, we suggest the design methodology of high-performance Lambda network, which can integrate NRPS (Network Resource Provisioning System) into our modified DRS model.

• Journal of KIISE:Computer Systems and Theory
• /
• v.34 no.9
• /
• pp.425-434
• /
• 2007

Even though Internet backbone speeds have increased in the last few years due to projects like Internet 2 and NGI, many high performance distributed applications are able to achieve only a small fraction of the available bandwidth. The cause of such problem is due to a character of TCP/IP. The primary goal of this protocol is reliable data transmission. Therefore high speed data transmission didn't be considered when TCP/IP is designed. Hence several researchers have been studied in order to solve the problem of TCP/IP. One of these research results, parallel transfer technique, solves this problem to use parallel TCP connections on application level. Additionally, this technique is compatibility. Recently, these researchers have been studied a mechanism to decide the number of parallel TCP connections. However, some researchers reported the number of parallel TCP connection base on only empirical results. Although hardware performance of host affects transmission rate, the hardware performance didn't be considered in their works. Hence, we collect all data related to transmission rate, such as hardware state information (cpu utilization, interrupt, context switch). Then, we analyzed collected data. And, we suggest a new mechanism determining number of parallel TCP connections for maximization of performance based on our analysis.