Yongqi He

Drop policies and multiple edge thresholds to enhance the performance of TCP over OBS networks with retransmission

Due to the bufferless nature of OBS networks, contention will unavoidably occur in core nodes. When the retransmission scheme is introduced into OBS networks, the bursts in the contention will be dropped selectively according to the drop policy adopted in core nodes, and retransmitted by edge nodes. In this paper, we proposed a drop policy based on Hop Number, in which the bursts with the larger total hop number will be given the higher priority in the contention. Taking the retransmission cost into account, we proposed another novel drop policy based on Hop Number Factor. Meanwhile, in edge nodes, multiple thresholds are also proposed to limit the unnecessary retransmission, and ease the deteriorated network performance. Both theoretical analyses and simulation results show that the proposed mechanisms can reduce the burst loss rate and the round trip time effectively, and improve the throughput significantly in TCP over OBS networks with retransmission.

Combining core drop policy and edge determinant threshold in TCP over OBS networks with retransmission

We proposed a novel drop policy in the core nodes which is combined with the determinant strategy in the ingress edge nodes. The proposed drop policy is based on the field of Hop Number (HN) taken by the burst control packets, which is introduced to determine which burst should be dropped when the contention happened in the core nodes. In the drop policy, the long-hop traffic is given the high priority, and most of the retransmitted traffic is left to be short-hop traffic. Therefore, there is a trade-off between the short-hop traffic and the long-hop traffic. The determinant strategy in the edge nodes is an initialized threshold, Retransmission Number Threshold (RNT), which is introduced to determine whether to start a retransmission operation when NAK is received. The unnecessary retransmissions in the network are limited, and the burst loss rate is reduced. The mechanism also takes the upper layer, TCP layer, into account. When the network has already been in the state of real congestion, the retransmission will only deteriorate the network performance. In the case, the combined mechanism leaves the retransmission process to the TCP layer. It can improve the network performance cost-effectively.

A novel node architecture for all-optical switching networks

The explosion of Internet traffic has brought about an acute need for high-performance networks. The bandwidth available on a single fiber has increased dramatically by the wavelength division multiplexing (WDM) technologies. How to use the huge bandwidth flexibly and efficiently is one of the hot topics of optical networks. It is high time that the high-performance switches/routers should be designed. All-optical switching network is a networking platform with effective and agile utilization of the available optical bandwidth. Some promising optical switching networks, such as optical packet switched, optical burst switched, and optical label switched networks are proposed. The benefit of these networks compared to the existing optical WDM networks rises from the higher network utilization at sub-wavelength granularity and from the supporting of more various services. However, there are still a lot of challenges. One challenge of them is contention. Due to the fact that the viable optical random access memory is not available in the optical domain so far, there exists a real possibility that packets may contend with one another at a switching node. In this paper, a novel node architecture called packet calking switch is proposed for all-optical switching networks. The proposed architecture is accompanied by a signaling protocol for packet-based traffic handling. This proposed node architecture is introduced into a simple network and a mesh network through simulation. The simulation results show that it is suitable for prioritized packet transmission. And it is more cost-effective than the existing node architectures because it requires much smaller optical switches and wavelength conversion to achieve nearly the same packet drop performance as the node configured partial wavelength conversion.

TCP performance analysis of grid over OBS networks

TCP performance is analyzed in Grid over OBS network with the high Bandwidth-Delay Product scenario. The impacts of the influential factors of both the TCP layer and the OBS layer are analyzed all-roundly.

p-cycle-based strategy for adaptive PWCE design

Future WDM optical networks are foreseen to be more dynamic and sensitive to failures. Therefore, dynamic service provisioning with network survivability becomes a critical requirement for network planning and management. The concept of protected working capacity envelope (PWCE) is attractive because it does not need protection resources to be dynamically configured thereby simplifying network management and reducing processing overheads. In this paper, we consider a p-Cycle-based Strategy for adaptive PWCE (protected working capacity envelope) design where PWCE can adaptively change with dynamically changing traffic. The blocking performance of this new strategy is compared to that of p-Cycle-based PWCE under a dynamic and non-uniform traffic model. The results show that the new strategy can adapt well to dynamic traffic demands that are non-uniform among the network nodes.

Drop Policy Combined with Multiple Edge Thresholds in TCP over OBS Networks

In this paper, we proposed a novel Burst Drop Policy based on Hop Number Factor (HNF) in the core nodes of TCP over OBS networks, which is combined with Multiple Thresholds in the edge nodes. The proposed drop policy, HNF, takes the retransmission cost into account, and gives the burst with the larger Hop Number Factor the higher priority in the contention. On the other hand, for the retransmission is not always effective, multiple Edge Thresholds, Retransmission Number Threshold (RNT), Traffic load Threshold (TlT) and Total hop Threshold (ThT), are proposed to limit the unnecessary retransmission in the OBS layer. If any of the thresholds is exceeded, the retransmission will be handed over to the upper layer, the TCP layer, which will be more cost-effective. The simulation results show that the combined mechanism can reduce the burst loss rate, decrease the round trip time and improve the throughput of TCP over OBS networks significantly.

Flow-oriented Dynamic Assembly Algorithm in TCP over OBS Networks

OBS is envisioned as a promising infrastructure for the next generation optical network, and TCP is likely to be the dominant transport protocol in the next generation network. Therefore, it is necessary to evaluate the performance of TCP over OBS networks. The assembly at the ingress edge nodes will impact the network performance. There have been several Fixed Assembly Period (FAP) algorithms proposed. However, the assembly period in FAP is fixed, and it can not be adjusted according to the network condition. Moreover, in FAP, the packets from different TCP sources are assembled into one burst. In that case, if such a burst is dropped, the TCP windows of the corresponding sources will shrink and the throughput will be reduced. In this paper, we introduced a flow-oriented Dynamic Assembly Period (DAP) algorithm for TCP over OBS networks. Through comparing the previous and current burst lengths, DAP can track the variation of TCP window, and update the assembly period dynamically for the next assembly. The performance of DAP is evaluated over a single TCP connection and multiple connections, respectively. The simulation results show that DAP performs better than FAP at almost the whole range of burst dropping probability.

Changes of traffic characteristics after large-scale aggregation in 3Tnet: modeling, analysis, and evaluation

Understanding network traffic behavior is essential for all aspects of network design and operation, e.g. component design, protocol design, provisioning, operations, administration and maintenance (OAM). A careful study of traffic behavior can lead to improvements in underlying protocols to attain greater efficiencies and higher performance. Many researches have shown that traffic in Ethernet and other networks, either in local or wide area networks, exhibit properties of self-similarity. Several empirical studies on network traffic indicate that this traffic is self-similar in nature. However, the network modeling methods used in current networks have been primarily designed and analyzed under the assumption of the traditional Poisson arrival process. These Poisson-like models suggest that the network traffic is smooth, which is inherently unable to capture the self-similar characteristic of traffic. In this paper, after introduce the high performance broadband information network (3Tnet) of China, an aggregation model at access convergence router (ACR) is proposed and analyzed in 3Tnet. We studied the impact of large-scale aggregation applied at the edge of 3Tnet in terms of the self-similarity level observed at the output traffic in presence of self-similar input traffic. Two formulas were presented to describe the changes of Hurst parameter. Using OPNET software simulator, changes of traffic characteristics after large-scale aggregation in 3Tnet was extensive studied. The theoretic analysis results were consistent with the simulation results.

IP calking: a novel decrease contention scheme in optical burst switched networks

Optical burst switching (OBS) is thought to be the best way to adapt the bursty IP traffic over optical wavelength division multiplexing (WDM) networks. OBS technology facilitates the efficient integration of both IP and WDM, and it provides statistical multiplexing gains and avoids long end to end setup time of traditional virtual circuit configuration. However, burst contention is inescapable in OBS for connectionless transmission and the absence of optical random-access memory. So many contention resolution methods are proposed. In this paper, a novel decrease contention scheme--IP calking was proposed for optical burst switched networks. IP calking method uses IP packets to carry single-hop traffic and the burst to carry multihop traffic as well, filling gaps between bursts by stuffing IP packets between adjacent nodes. A statistical analysis model was developed to analyze the performance of this method. In an arbitrary network, the analyses indicate that the decrease of packet dropping probability is directly proportional to the link number and almost inversely proportional to the square of the node number. The theoretical results are validated through extensive simulations. Simulation results show that IP calking decreases about 50 % data drop probability of the no calking scheme and the link utilization is improved 5~25% which varying with the offered load in a simple network.

Group scheduling based on control-packet batch processing in optical burst switched networks

Optical burst switching (OBS) is proposed as a high-speed, flexible, and transparent technology. It is thought to be the best way to adapt the bursty IP traffic over optical wavelength division multiplexing (WDM) networks. OBS technology facilitates the efficient integration of both IP and WDM. It provides statistical multiplexing gains and avoids long end to end setup time of traditional virtual circuit configuration. However, there are still a lot of challenges, one of which is burst contention. Owing to the fact that random access memory like buffering is not available in the optical domain at present, there exists a real possibility that bursts may contend with one another at a switching node. Many contention resolutions are proposed. The major contention resolutions in literature are wavelength conversion, fiber delay lines, and deflecting routing. In this paper, a new data burst scheduling scheme, called group scheduling based on control-packet batch processing (GSCBP) was proposed to reduce burst contention. Like transmission control protocol, GSCBP has a batch processing window. Control packets which located in the batch processing window are batch processed. A heuristic scheduling algorithm arranges the relevant bursts route based on the processing result and the network resource. A new node architecture supporting group scheduling was presented. The GSCBP algorithm is combined with wavelength converter and/or fiber delay lines which is shared by some data channels. Meanwhile, an extended open shortest path first (E-OSPF) routing strategy was proposed for OBS. Both GSCBP and E-OSPF are introduced into 14-node national science foundation network by means of simulations. The ETE delay, burst blocking probability, as well as burst dropping probability were attained. Results show that the GSBCP lead to the higher-priority traffic drop rate decrease one order of magnitude, if drop rate and ETE delay of lower priority traffic is sacrificed.