• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.8 no.3
• /
• pp.762-777
• /
• 2014

With the increasing usage of cloud applications such as MapReduce and social networking, the amount of data traffic in data center networks continues to grow. Moreover, these appli-cations follow the incast traffic pattern, where a large burst of traffic sent by a number of senders, accumulates simultaneously at the shallow-buffered data center switches. This causes severe packet losses. The currently deployed TCP is custom-tailored for the wide-area Internet. This causes cloud applications to suffer long completion times towing to the packet losses, and hence, results in a poor quality of service. An Explicit Congestion Notification (ECN)-based approach is an attractive solution that conservatively adjusts to the network congestion in advance. This legacy approach, however, lacks scalability in terms of the number of flows. In this paper, we reveal the primary cause of the scalability issue through analysis, and propose a new congestion-control algorithm called FaST. FaST employs a novel, virtual congestion window to conduct fine-grained congestion control that results in improved scalability. Fur-thermore, FaST is easy to deploy since it requires only a few software modifications at the server-side. Through ns-3 simulations, we show that FaST improves the scalability of data center networks compared with the existing approaches.

• The KIPS Transactions:PartC
• /
• v.8C no.3
• /
• pp.335-344
• /
• 2001

최근 인터넷 환경이 반도체, 광통신 그리고 차세대 인터넷 기술의 발달로 고성능화 되어가고 있다. 따라서 고성능 인터넷을 위한 TCP의 성능 향상 연구가 매우 중요해졌다. 그러나 기존 TCP는 수신위도 버퍼의 물리적 크기에 의하여 최대 전송 성능과 대역폭 탐색 기능이 제한을 받는 구조적인 문제점을 갖고 있다. 본 논문에서는 이를 해결하기 위하여 수신 호스트에 가상 윈도 개념을 도입하였다. 이는 송신 호스트가 RTT 동안 균일하게 세그먼트를 분산시켜서 패킷을 전송할 때 세그먼트 간격 시간 동안 수신 호스트의 처리 능력을 가상윈도로 나타내는 것이다. 따라서 가상 윈도의 크기는 수신 호스트의 성능에 비례하기 때문에 수신 호스트가 고성능일 경우 TCP의 전송 능력 성능이 더 높아질 수 있다. 초고속 인터넷일 경우 제안 알고리즘이 기존 TCP보다 전송능력에 있어 1.5∼5배 개선되는 것을 네트워크 시뮬레이션인 NS2를 이용하여 확인하였다.

• Journal of Communications and Networks
• /
• v.4 no.2
• /
• pp.148-157
• /
• 2002

This paper investigates schemes to improve TCP performance in multipath forwarding networks. In multipath routing, packets to the same destination are sent to multiple next-hops in either packet-level or flow-level forwarding mode. Effective bandwidth is increased since we can utilize unused capacity of multiple paths to the destination. In packet-level multipath forwarding networks, TCP performance may not be enhanced due to frequent out-of-order segment arrivals at the receiver because of different delays among paths. To overcome this problem, we propose simple TCP modifications. At the sender, the fast retransmission threshold is adjusted taking the number of paths into consideration. At the receiver, the delayed acknowledgment scheme is modified such that an acknowledgment for an out-of-order segment arrival is delayed in the same way for the in-order one. The number of unnecessary retransmissions and congestion window reductions is diminished, which is verified by extensive simulations. In flow-level multipath forwarding networks, hashing is used at routers to select outgoing link of a packet. Here, we show by simulations that TCP performance is increased in proportion to the number of paths regardless of delay differences.