Transmission control for heterogeneous network composed of MANET and terrestrial-satellite network

Abstract

The satellite network has the advantages of wide coverage, convenient access, and the influence of the natural environment. Wireless multi-hop network is suitable for the occasion that cannot be easily or inconvenient to lay network facilities and the occasion of the need for rapid automatic networking. Multi-hop network and the satellite network may combine a heterogeneous network that can provide an effective and convenient way to access the network in the ocean or in remote areas. This paper, firstly, builds a heterogeneous network with satellite network and multi-hop network, and discusses the bandwidth delay product (BDP) of heterogeneous network, then proposes a new end-to-end congestion control algorithm. The proposed scheme at the outset increases the dynamic time interval for the initial transmission data that can reduce the backlog of burst data at the access point, and then increases the amount of data transmission in the slow start that can improve high bandwidth delay product of satellite network. In the congestion avoidance, it adjusts the congestion window size by monitoring the random packet loss and congestion packet loss. This algorithm can effectively solve the long delay in heterogeneous networks, and improve the accuracy of the high packet loss rate monitoring in multi hop networks and satellite networks. The proposed algorithm can effectively solve the influence of the long delay in the heterogeneous network, at the same time it can distinguish the different packet loss.

Keywords

Congestion control multi-hop network satellite network heterogeneous networks

1 Introduction

The modern network showed a trend of highly heterogeneous. Wireless Local Area Networks (WLAN), 3 G and 4 G mobile communication terminals connected to the wired network greatly improved the convenience of the network access for people. But in the absence of infrastructure such as the remote areas and ocean, satellite network has become the best access choice. As a supplement to the ground network, satellite network is an important part of the Global Internet. It has its own characteristics, such as long propagation delay, high link error rate and bandwidth asymmetry, etc.

Wireless multi-hop network is suitable for the occasion of unable or inconvenient to lay network facilities and the occasion of the need for rapid automatic networking. It can be supported the wireless transmission of data, voice, image and other services in the bad environment through temporary networking. Wireless multi-hop network has the characteristics such as distributed control, self-organizing, dynamic topology, limited bandwidth, limited energy, multi-hop routing and so on. The ground of the wireless multi-hop network can be connected with an external network through the satellite network. This heterogeneous network that consisting of satellite networks and multi hop networks is particularly suitable for military battlefield, offshore and remote areas.

The network congestion is a focus point in the heterogeneous network. Satellite network and wireless multi-hop network have own characteristics for the congestion control. The current TCP (New Reno) suffers performance problems in geostationary satellite networks due to prolonged propagation delays and high bit error rates. The International Standardization Organization such as the Consultative Committee for Space Data Systems (CCSDS), proposes transmission protocol SCPS-TP [1]; Internet Engineering Task Force (IETF) works on the RFC documents [2, 3]. In recent years, many scholars propose a large number of solutions for improving TCP performance in satellite networks [4, 5]. The proposed solutions can be divided into the following four categories: speed up the transmission for the long propagation delay [6], modify satellite network link bandwidth asymmetry or high bit error rate (BER) [7], and apply Delay/Disruption Tolerant Networking (DTN) to space communications [8, 9]. Gupta and Goyal applied a fuzzy based algorithm for finding the congestion free shortest path and increased the performance of the network [10]. Conti and Giordano analyzed four successful networking paradigms, mesh, sensor, opportunistic, and vehicular networks, that emerged from the MANET world as a more pragmatic application of the multihop ad hoc networking paradigm. They also presented the new research directions in the multihop ad hoc networking field: peoplecentric networking, triggered by the increasing penetration of the smart phones in everyday life, which was generating a people-centric revolution in computing and communications [11]. Mbarushimana et al. first evaluated the performance of TCP in 802.11e MANETs when competing with high priority VoIP traffic, and then proposed a novel TCP-friendly scheme, called IEDCA, to improve IEEE 802.11e EDCA mechanism [12]. Oliveira et al. used a topology control algorithm that increased the path duration by decreasing the probability of path breaks for a high-speed highway mobility scenario [13]. Chowdhury et al. investigated the limitations of classical TCP newReno in a cognitive radio ad hoc network environment, and proposed TCP CRAHN, incorporated spectrum awareness by a combination of explicit feedback from the intermediate nodes and the destination [14]. Jehan et al. described the performance behavior of BIC, Cubic, TCP Compound, Vegas, Reno and Westwood congestion control algorithms on MANETs [15]. Lochert et al. given an overview over existing proposals, explained their key ideas and showed their interrelations [16].

The heterogeneous network congestion control has been attracting the attention of the researchers. Vijayalakshmy et al. proposed a hybrid coupled interworking of three networks (WiMAX – WLAN – LTE – IPQoS – LB) using H. 323 signaling protocol. Heterogeneous network model based on Fast handover Hierarchical Mobile IPV6 (FHMIPV6) protocol that integrates the WiMAX, LTE and WLAN technologies is proposed to improve QoS [17]. Li et al. proposed a new multipath TCP protocol, namely SC-MPTCP, by integrating linear systematic coding into MPTCP. It used of coded packets as redundancy to counter against expensive retransmissions [18]. Shukla and Tyagi studied about the network mobility approach in vehicular ad hoc network; the model described the movement of vehicles from one network to the other network. The proposed handover scheme reduces the handover latency, packet loss signaling overhead [19]. Li designed TCP congestion control algorithm based on model bandwidth estimation to solve the problem of heterogeneous network switching. The results validated the effectiveness of this mechanism under simulation analysis, which effectively achieved the fair competition of resource and can improve the utilization of network resources [20]. Kim et al. proposed reordering considered packet scheduling algorithm at the sender. Main idea of the proposed scheme was estimating packet receiving time at the receiver and scheduling packet. By using the proposed algorithm, the number of out-of-ordered packet in the receiver buffer was reduced and the performance degradation was also mitigated especially in the case that asymmetry between two path was severe [21]. Yew et al. investigated the performance of TCP Vegas in both homogeneous and heterogeneous wired and wired-cum-wireless networks. The performance of TCP Vegas was the worst compared to other TCP variants in homogeneous wired-cum-wireless network. TCP Vegas had a significantly lower delay than other TCP variants in both homogeneous and heterogeneous wired and wireless networks [22]. Hu et al. proposed a two-layer hierarchical cache architecture for enhancing TCP performance over heterogeneous networks with both wired and wireless links. A new network-layer protocol, called new snoop (NS), is designed. The main idea is to cache the unacknowledged packets at both the mobile switch center (MSC) and base station (BS), to form a two-layer cache hierarchy [23]. Barakat et al. studied TCP performance independent of the type of network by considering the different possible characteristics of the connection path. They presented the problems and the different proposed solutions. This study permitted us to understand the limitations of the actual solutions and the required modifications to let TCP cope with heterogeneous Internet on an end-to-end basis [24].

2 The heterogeneous network model composed of multi-hop networks and satellite networks

2.1 The model of the heterogeneous network

The general heterogeneous network mainly refers to the wired network and the wireless network, such as the heterogeneous network composed of satellite network and terrestrial network. The analysis of heterogeneous network composed of satellite networks and multi-hop networks is involved little. Wireless multi-hop networks are adapted to the absence of infrastructure and emergency situations, and satellite networks can be connected to anywhere in the world. The construction of heterogeneous network can be further extended scope of wireless network communication.

As shown in Fig. 1, the wireless multi-hop network can be Ad Hoc network, mobile Ad hoc network (MANET), Mesh network or wireless sensor network (WSN). For wireless multi-hop networks, it can be only internal communication. This is the same as the nature of the independent wireless multi hop network. When the wireless multi-hop networks need communication with the external network, it can be connected to the network via wireless local area network, cellular network, and satellite network etc. Since the initial design of the wireless multi-hop network is to design the lack of infrastructure, the use of satellite network is a very reasonable supplement to the wireless multi-hop network. This new heterogeneous network is a great expansion and a new challenge to the existing network communication, which is suitable for the emergency situations of marine communications, remote areas, military battlefield and disaster.

Fig.1

Heterogeneous network model composed of multi-hop networks and satellite networks.

In the wired network, the network technology is quite mature, and there are a lot of infrastructures for communication network service, so it has great advantages both the available bandwidth and the bandwidth delay product. The main factors that affect the performance of heterogeneous networks are satellite networks and multi hop networks. In the satellite network, it has long propagation delay (GEO satellites network round-trip time can reach 550 ms), as well as today’s increasingly wide application of broadband satellite network. Satellite network bandwidth delay product (BDP) has a larger value.

The bandwidth of the physical channel of wireless multi-hop networks is obviously cannot be compared with the wired network, as well as its round-trip time generally less than at the round-trip time of the satellite network. When data transmission has experienced a transmission time of the one-way satellite link reach to the access point, the data forwarding from the access point to the multi-hop network has been completed, or loss due to the overflow of the buffer (satellite network burst data transmission). So the long propagation delay of the satellite network is a key factor affecting the performance of the whole heterogeneous network. Then the high bit error rate of the satellite network will cause a lot of data lose. The number of data for retransmission will increase. Due to the hidden terminal and exposed terminal problem in wireless multi-hop network, if the back off time is too long will cause data loss when listen to the transmission channel.

2.2 The bandwidth delay product of the heterogeneous network

In Fig. 2, the heterogeneous network is composed of a chain of wireless multi-hop networks and terrestrial-satellite networks. The bandwidth delay product from the wired network to the satellite network, and finally to the multi-hop network is gradually reduced. In order to obtain the maximum throughput of the network, the volume of transmission data is close to the bandwidth delay product of the bottleneck network.

Fig.2

The bandwidth delay product of Heterogeneous network.

The bandwidth delay product of satellite network is BDPs = Bandwidth×Delay. The bandwidth delay product of an N-hops chain wireless multi-hop network will not exceed ${BDP}_{Mupper} \leq \frac{\sum_{i = 0}^{m} d_{i} + \sum_{i = 0}^{m} d_{i}^{'}}{4 d_{max}}$ (1)

Where d_i and d′ are the per-hop packet transmission delays along the forward and return paths, d_max is the maximum per-hop delay [25].

When the number of hop-count increases, the bandwidth delay product is linear increase in the chain multi-hop network. Theoretically, the bandwidth delay product of the chain multi-hop network is no longer the performance bottleneck of the whole heterogeneous network. As shown in Fig. 2, at this point the satellite network becomes the bottleneck of the whole network. When the bandwidth delay product of the chain multi-hop network is less than the bandwidth delay product of the satellite network, the long propagation delay of the satellite network is a key factor affecting the whole heterogeneous network. So it has the following theorem.

Theorem. In the heterogeneous network that is composed of the chain of wireless multi-hop networks and satellite networks, the bandwidth delay product of the chain multi-hop network will exceed the bandwidth delay product of the satellite network when the number of hop-count increases. $lim_{m \to \infty} \frac{\sum_{i = 0, i \neq max}^{m} d_{i} + d_{max} + \sum_{i = 0}^{m} d_{i}^{'}}{4 d_{max}}$ (2)

In all the hops in the chain multi-hop network, when ∀m = max, it has a d_max, d_max → C₂.

The above formula can be simplified asfollowing:

$\begin{matrix} lim_{m \to \infty} \frac{\sum_{i = 0, i \neq max}^{m} d_{i} + \sum_{i = 0}^{m} d_{i}^{'}}{4 C_{2}} + \frac{1}{4} \\ = lim_{m \to \infty} \frac{1}{4 C_{2}} \sum_{i = 0, i \neq max}^{m} d_{i} + \frac{1}{4 C_{2}} \sum_{i = 0}^{m} d_{i}^{'} + \frac{1}{4} \end{matrix}$ (3)

So $lim_{m \to \infty} {BDP}_{Mupper} (m) \to \infty$ (4)

Obviously, with the increase of m, the forward and backward delays in the chain multi-hop network tend to infinity, and the bandwidth delay product of the chain multi-hop network tends to infinity.

Remark. The bandwidth delay product of the wireless multi-hop network isn’t exceed the AB_upper ${AB}_{upper} = {\begin{matrix} \frac{C}{n}, & 1 \leq n \leq 4 \\ \frac{C}{4}, & n > 4 \end{matrix}$ (5)

Where C is maximum physical channel rate, n is the path number of hops [26, 27]. If the wireless multi-hop network takes IEEE802.11b which the physical channel rate is 11 Mb/s, the available bandwidth of the wireless multi-hop network is 2.7 Mb/s when hop-counts exceed 4 hops. And the throughput of the TCP Reno with satellite network is: $Throughput = \frac{{Win}_{max}}{RTT}$ (6)

TCP Reno’s Win_max is maximum transmission window that is set 64 KB. In the GEO satellite network, the round-trip time can reach 550 ms, the maximum throughput of the satellite link will not exceed 1 Mb/s. This shows that improving the performance of the satellite network can improve the performance of the whole heterogeneous network.

3 Congestion control algorithm for the heterogeneous network

Congestion control has always been a hot area of research in heterogeneous networks [28, 29]. Congestion control for heterogeneous networks is often used to performance enhancing proxies (PEPs). It splits the network protocol stack into several TCPs to improve the performance of the heterogeneous network, especially in the terrestrial-satellite network. However, in the heterogeneous network that is composed of multi-hop networks and satellite networks, if the terminal network in heterogeneous networks frequently moving, the frequent mobility can cause the problems of switching and synchronization between multi-hop nodes and gateway [30]. The frequent switching, in addition to increasing energy consumption dramatically for energy-efficient multi-hop network (such as wireless sensor network), but also increases the additional transmission delay. This paper proposes an end to end scheme that includes dynamic time interval and burst data and control.

3.1 Dynamic time interval

Satellite networks and multi hop networks have great differences in the amount of data transmitted. The satellite networks need to increase the amount of data to fill the larger bandwidth delay product, while in the multi hop network, it is generally believed that the smaller data transmission amount is beneficial to reduce the channel competition and improve the performance of the network. When increasing the data transmission of the satellite network, there is a large amount of data backlog in the gateway between the satellite network and multi-hop network, which can easily lead to congestion and loss of data.

From analysis of Section 2.2, in order to make use of the high bandwidth delay product of satellite networks, the heterogeneous network must improve the data transmission of satellite networks. For the burst data of satellite networks, the appropriate interval time can effectively alleviate the data backlog at the gateway.

This paper proposes a scheme that uses a dynamic time interval in the initial phase to ease the burst of data transmission, and the dynamic time interval T_n is set as follows: $T_{n} = \frac{| {RTT}_{n} - {RTT}_{min} | + | {RTT}_{n} - {RTT}_{n - 1} |}{2 \times {Count}_{CW}}$ (7) where RTT₀ is the initial round-trip time, which is obtained by the initial TCP connection, and Count_CW is the number of packets sent.

The initial value of RTT_min is set up as a one-way propagation delay in satellite networks. When the difference between two consecutive round-trip times |RTT_n - RTT_n-1| increases, that shows the heterogeneous network state changes dramatically, and the dynamic interval time is increased. If two consecutive round-trip times are measured in the congestion state, the |RTT_n - RTT_n-1| will lead to a smaller difference and the dynamic interval time will be reduced. While the value of |RTT_n - RTT_min| can effectively detect the transmission status of the heterogeneous network, this can avoid data transmission frequency too fast. When the number of packets is increased, the dynamic time interval decreases. This will not affect the transmission efficiency of long-LiveTCP.

Undoubtedly, increasing the number of time intervals can effectively avoid the loss of packets in the gateway that connects the satellite network and multi-hop network. However, if the time intervals are too long, it will reduce the efficiency of the heterogeneous network for Long-Live TCP [30]. For most of application services, data transfer can be completed in the first few round-trip times. So this paper uses the first 3 round-trip times to carry out the dynamic time interval processing. For Short-Live TCP application services, data transfer can be completed in three round-trip times, and our scheme has the minimal influence for long-Live TCP.

3.2 Burst data and its control

There are many improved schemes of TCP for satellite network. This paper refines TCP Hybla [31], which is end-to-end scheme, increases the amount of data transmission uses the ratio of the long delay of the satellite network and wired delay. This is very beneficial for Long-Live. TCP

3.2.1 TCP Hybla scheme

To eliminate penalization of TCP connections that incorporates a high-latency satellite radio link with longer RTT. TCP Hybla modifies the congestion control rules with the aim of obtaining the same instantaneous segment transmission rate for long RTT connections as a comparatively fast reference TCP connection (e.g. a wired one). To this end, a normalized round trip time, ρ, was introduced as the ratio between the actual RTT and the round trip time of the reference connection, denoted by RTT₀. $ρ = \frac{RTT}{{RTT}_{0}}$ (8)

Such a nomalization helps to remove the performance dependence on RTT. Through a theoretical proof, the same segment transmission rate of the reference connection can be achieved by longer RTT connections by replacing the congestion control rules after receiving a new ACK with the following, $W_{i + 1} = {\begin{matrix} W_{i} + 2^{ρ} - 1, SS phase \\ W_{i} + ρ^{2} / W_{i}, CA phase \end{matrix}$ (9) where W_i denotes the cwnd with the reception of the ith ACK.

To deal with the issues of long RTT and high BER over GEO satellite network at the same time, we propose an enhanced TCP scheme by combining the benefits of Hybla and Veno. Specifically, our proposal enhances the slow start phase of TCP Hybla to mitigate the effect of long RTT, and refines Veno’s fast recovery algorithm to distinguish between random packet loss and congestion loss. In this Section, we describe the proposed enhanced TCP scheme in the slow start and fast recovery algorithms.

3.2.2 Congestion window increase algorithm in slow start

In our proposal, the cwnd update equations of TCP Hybla in (9) are still employed in the SS and CA phases. However, the setting of RTT₀ is modified and the congestion window increase algorithm is refined in the SS phase.

In Hybla, the update rule of cwnd uses a parameter ρ, which is the ratio of actual RTT compared with a reference round trip time RTT₀. The value of RTT₀ is the round trip time of wired connection, and chosen to be RTT₀ = 25 ms for evaluating Hybla’s performance. However, the RTT of GEO satellites can reach 550 ms, and ρ can be up to 22. Consequently, the update of the congestion window can be up to 2²² –1 (4,194,303), and this value is too large for the congestion window in the SS phase. Park et al. suggests the value of RTT₀ takes 70 ms [32] to verify the performance enhancement with packet loss over the satellite network. Considering the existence of congestions, we suggest to set it to 75 ms. Thus, the maximum value of ρ becomes 7, and the update of congestion window can reach 2⁷ –1 (127).

In addition, the SS phase of TCP is only decided based on cwnd and ssthresh. When both congestion and random packet losses exist in the network, such a decision may not reflect the real situation of the network. Thereby, we propose a new congestion window increase mechanism for cwnd in the SS phase. The new mechanism divides the SS phase into three sub-phases, which increases the sent rate taking into account of cwnd, ssthresh, rwnd (receiver advertised window), and flightsize (total unacknowledged byte in the network). Thus, the congestion window increase can be adapted according to the different network situations in the SS phase. The division of the sub-phases and the corresponding update rules of cwnd are as follows,

If (cwnd< ssthresh) //SS phase

{

if (cwnd≤ssthresh/4) & & (flightsize < w nd/4)

//sub-phase 1

set cw nd = 2 * W_i₊₁

if (ssthresh/4 < cwnd < ssthresh3/4)

& & (rwnd/4≤flightsize < rwnd3/4) //sub-phase 2

set cwnd = W_i₊₁

if (cwnd≥ssthresh3/4) || (flightsize≥rwnd3/4)

//sub-phase 3

set cwnd =W_i+1/2

}

In the sub-phase 1, the values of cwnd and flightsize are very small, which means that there are not much sent and unacknowledged data in the network. Therefore, the sent rate can be increased aggressively as cwnd = 2 * W_i₊₁. In the sub-phase 2, the values of cwnd and flightsize are in the intermediate range. In this sub-phase, the update rule of congestion window is chosen to be the same as TCP Hybla (i.e., cwnd = W_i +1). In the sub-phase 3, the cwnd approaches ssthresh or the flightsize approaches rwnd. It is more likely that large backlogs exist in the network. In order to maintain the network stability, the sent rate should be less aggressive, and cwnd is set to W_i+1/2.

In the SS phase, our proposal employs the cwnd growth rate of Hybla, and combines cwnd, flightsize, ssthresh, and rwnd four variables to determine the increase rate of congestion window. It intends to update the congestion window adaptively corresponding to different network situations to improve the TCP performance and maintain the stability of the network.

3.2.3 Congestion window decrease algorithm in fast recovery

As we mentioned, traditional TCP variants (Tahoe, Reno, NewReno, etc.) treat random packet loss as the congestion loss, which results in TCP performance degradation; on the other hand, TCP Veno [33] can distinguish effectively between error packet loss and congestion loss in the wireless environment. In Veno, the value of β is set 3 for distinguishing between random loss and congestion loss. Such a setting may not be appropriate for the long delay and high BER satellite link. Moreover, the reasons for packet loss are two-level differentiations (either random loss or congestion loss) according to the value of backlog data in Veno. For the high BER satellite links, it may not be so decisive on the reasons for packet losses just based on the backlog data. Thus, we proposed to refine the congestion window decrease algorithm in the fast recovery algorithm. When the sender receives three DUPACKs, the Multiplicative Decrease algorithm in our proposal is modified as follows,

if (N < β1) // random loss is most likely to have occurred

set ssthresh = cw nd * 4/5;

else if (β1≤N < β2) // not certain which kind of loss else if loss occurred

set ssthresh = cw nd * 3/5;

else set ssthresh = cw nd * 2/5; // congestion loss is most likely to have occurred

In our simulations, β₁ and β_2are set to be 3 and 6, respectively. In our proposal, three conditions of N are considered. When N ≤ 3, ssthresh takes Veno’s setting, 4/5 * cw nd. When 3≤N < 6, set ssthresh to 3/5 * cw nd; For N≥6, set ssthresh to 2/5 * cw nd. Though the last threshold value is smaller than the traditional fast recovery value 1/2 * cw nd, the congestion window is still enough for the transmission.

In our proposal, different thresholds are set corresponding to the value of backlog data, which make multiple-level differentiations on the network conditions. Comparing the original Veno’s algorithm, multi-level threshold adjustment in our proposal gives refined distinguishing of congestion loss and error loss, which is more reasonable for the high BER satellite links.

Apart from the above enhancements of the congestion control rules, some mechanisms employed in Hybla are maintained in our proposal, including the adoption of the SACK policy and the use of timestamps.

4 Performance analysis

The topology of simulation is shown in Fig. 3. The satellite network connects with the wired network and the wireless multi-hop network. In the multi-hop network, each node downloads a file from the server in the wired network. The date transmission travel through to the wired network, satellite network and finally reach to the each node in the multi-hop network. In the satellite networks, the link bandwidth is set 1.54 Mbps, and the one-way propagation delay is set 250 ms. The bit error rate (BER) of the satellite link changes from 10^–9 to 10^–5. Wireless multi-hop network is composed of a chain of 5 hops. The nodes take IEEE 802.11 standard in the multi-hopnetwork.

Fig.3

The topology of simulation.

The simulation for performance evaluation is conducted using OPNET 14.5. In the heterogeneous network, it must meet requirements to reduce cost of satellite operation. The multi-hop network access to the satellite network requires the expensive cost. So the throughput of the satellite link is must considered. It reflects the performance of local heterogeneous network.

In the multi-hop network, each node downloads a 12.5 MByte file from the server. The simulation considers the download response time that elapsed between sending a request and receiving the response packet for the FTP application in this node. Measured from the time a client application sends a request to the server to the time it receives a response packet. This time includes the signaling delay for the connection setup and tear-down. This statistic is written after the connection is closed. The download response time reflects the performance of terminal nodes in heterogeneous networks. The simulation compared with the performance of four schemes: TCP Reno, TCP Veno, TCP Hybla and the proposed scheme.

In the wireless multi-hop network, the 5 nodes download a 12.5 MByte file from the server in the wired network, and the throughput of the satellite link is contrasted in the different hop counts, as shown in Fig. 4. When the BER of the satellite link is 10^–9, the throughput of the proposed scheme is higher than the other three kinds of TCPs. The main reason is that the proposed scheme makes full use of the BDP of the satellite link, and deals with differently in the process of the congestion packet loss and random packet loss. This greatly improves the throughput of the satellite link. In the four TCP schemes, the throughput of Reno TCP is lowest, and it maintains at a stable value.

Fig.4

The throughput of satellite link when the BER is 10^–9.

While comparing the throughput of TCP Hybla and TCP Veno can be showed that TCP Hybla can increase the amount of transmission data with a certain proportion of RTT of the satellite network and wired network, as the Section 2.2 analyses, it also improves the throughput in the satellite link.

However, it is also found that TCP Veno does not take TCP Hybla’s scheme that increases the amount of transmission data to improve the satellite throughput, and also does not take burst data to improve the transmission of data. But it distinguishes the packet random loss and congestion loss, and avoids reducing the amount of data transmission, which also indirectly improves the throughput on the satellite link.

Figure 5 is the throughput of satellite link when the BER is 10^–7. Compared with Fig. 4, the change range of the throughput of the four TCPs schemes is small. With the increase of the number of hop count, in addition to TCP Reno, the other three kinds of TCP’s throughput have a certain decline. The advantage of the proposed scheme is further expanded compare with TCP Hybla.

Fig.5

The throughput of satellite link when the BER is 10^–7.

When the BER of the satellite link is 10^–5, the throughput of the four TCP schemes on the satellite link is shown in Fig. 6. Compared with Figs. 4 and 5, the throughput has dramatically reduced. The advantage of the proposed scheme is further expanded compared with TCP Hybla and TCP Veno. When the hop count is just one hop, the proposed schemed does show higher the throughput of 22 kbps and 61 kbps than TCP Hybla and TCP Veno respectively, which benefit the proposed scheme not only increases the amount of transmission data, but also takes the dynamic time interval for the initial burst transmission data.

Fig.6

The throughput of satellite link when the BER is 10^–5.

The dynamic time interval alleviates the data backlog in the access points that connects the multi-hop networks, and reduces the data loss when the network occurs congestion. While the data random loss caused by high BER in the satellite link and congestion loss caused by buffer overflow in the intermediate nodes is more serious, the two factors are the main reasons causing the performance degradation of various TCP schemes. The proposed scheme can distinguish two kinds of data loss, and take correct the corresponding measures. At the same time, the data transmission is stable with high efficiency.

TCP Reno’s throughput of the satellite link is relatively stable when the hop count increase from the 1 hop to 5 hops. This isn’t obvious fluctuations and maintains 2.2×10⁴ bps. TCP Hybla is used to improve the bandwidth utilization of the satellite link by increasing the amount of data transmission under the long delay satellite network. TCP Veno efficiently distinguishes data random loss and congestion loss in heterogeneous network and reduces the data retransmit. These TCP schemes are suitable for the heterogeneous network composed of satellite networks and multi-hop network. As the Section 2.2 analysis, improving the performance of the satellite network can improve the performance of the whole heterogeneous network.

Figures 7 and 8 show the nodes download response time in the wireless multi-hop network, when the satellite link’s BER is 10^–9 and 10^–7, respectively. The download response time of TCP Reno is maintained at 4.8×10³s, and the fluctuation range is small. The download response time of TCP Hybla and TCP Veno is maintained between0.8×10³s and 1.2×10³s. With the increase of the number of hops, the download response has been increased. The proposed scheme’s download response time is least. Generally, when the BER of the satellite link is low, the heterogeneous network does data loss occurs, and the server of the wired network retransmit the loss data after waiting for a timeout.

Fig.7

Download response time when BER is 10^–9.

Fig.8

Download response time when BER is 10^–7.

The TCP Hybla sends a large amount of data to the heterogeneous network. Although it will cause a large amount of data backlog in the access point, but a large amount of transmission data offset the impact of the loss of data when the data loss is less. In contrast, although the amount of transmission data of TCP Veno is less than Hybla TCP, but it distinguishes the loss of data and avoids the rapid reduction of data transmission.

When the BER of the satellite link is 10^–5, the download response time of the four TCP schemes is shown in Fig. 9. Compared with Figs. 7 and 8, the download response time has dramatically increased. The download response time of the proposed scheme is least in the four TCP scheme. The difference of download response time between TCP Hybla and TCP Veno is being enlarged. TCP Reno’s download response time remains in a stable value from the 2 hops to 5 hops. With the increase of the number of hops in the wireless multi-hop network, the download response time of all TCPs is increased. The proposed scheme utilizes the advantages of TCP Hybla in satellite networks, and better utilizes the bandwidth of the satellite link, so as to improve the performance of the whole heterogeneous network.

Fig.9

Download response time when BER is 10^–5.

Figure 10 is queue delay of the nodes in the multi-hop network (including the access point). The statistic of the queuing delay represents instantaneous measurements of packet waiting times in the transmitter channel’s queue. Measurements are taken from the time a packet enters the transmitter channel queue to the time the last bit of the packet is transmitted. In the multi-hop networks, the queue delay of the access point (the first hop node in Fig. 10) is relatively large. This is due to the different transmission strategy of the satellite networks and multi-hop networks. In order to obtain greater throughput, the former is to increase the amount of the data transmission. The latter is to reduce the size of the data transmission. That is inevitable keep a large number of data backlog in the access point.

Fig.10

Queue delay of the nodes in the multi-hop network.

While with the increase of the hop count in the multi-hop network (from 2 hops to 5 in Fig. 3 corresponding 1 hop to 4 hops in the multi-hop network), the queue delay is gradually reduced. As the analysis in Section 2.2, with the increase of the hop count, the BDP of the nodes will be increases in the chain multi-hop networks. This can accommodate more data into multi-hop networks. Actually, the hop count is not infinite. Because of the characteristics of the multi-hop network and satellite network has huge difference. Therefore, the coordination of the data transmitted both the satellite network and multi-hop network will directly affects the performance of the whole heterogeneous network.

The proposed scheme adds a dynamic time interval for the initial data transmission, and avoids the congestion caused by the excessive data packets accumulation in the access points of the multi-hop networks. In practice, most of the application of data transmission can be completed in the first few RTT. It is very effective for the short-TCP connection. This not only eases the data backlog in the access point, but also it will not have too much influence for the long-TCP application.

5 Conclusion

The traditional heterogeneous network mainly is composed of the wired network and the wireless network. Research on heterogeneous networks that composed of satellite networks and wireless multi-hop networks is very sparse. Satellite networks can communicate with anywhere in the world, and the wireless multi-hop network adapts to the no infrastructure and emergency scenarios. Integrated satellite networks and wireless multi-hop networks can extend the communication range of the wireless network. But it also presents new challenges to the expansion of wireless communication networks.

In this paper, a new heterogeneous network is proposed, which composed of satellite network and wireless multi-hop network. By analyzing the BDP of the heterogeneous network, a congestion control scheme is proposed. The four kinds of TCP are simulated in the heterogeneous network. From the performance analysis, it is found that the transmission control protocol adapted to the satellite network has a great help to improve the performance of the heterogeneous network.

Footnotes

Acknowledgments

This work was supported by the key Scientific Research fun of Provincial Key Laboratory of Informational Service for Rural Area of South-western Hunan, the program for Science and Technology Innovative Research Team in Higher Educational Institutions of Hunan Province and the National Natural Science Foundation of China (Grant No. 61472135 and Grant No. 61561017).

References

CCSDS 714.0-B-2.” Space Communications Protocol Specification - Transport Protocol (SCPS-TP),” Blue Book Issue 2, Washington, DC, USA, 2006.

Allman

, Glover

and Sanchez

, Enhancing TCP over satellite links using standard mechanisms, RFC 2488, 1999.

Allman

, Dawkins

, Glover

, et al., Ongoing TCP research related to satellites. RFC 2760, 2000.

Taleb

, Hadjadj-Aoul

and Ahmed

, Challenges, opportunities, and solutions for converged satellite and terrestrial networks, IEEE Wireless Communications18(1) (2011), 46–52.

Botta

and Pescapé

, On the performance of new generation satellite broadband internet services, IEEE Communications Magazine52(6) (2014), 201–209.

Blasco

F.L.

, Rossetto

and Bauch

, Time interference alignment via delay offset for long delay networks, IEEE Transactions on Communications62(2) (2014), 590–599.

Sun

, Ji

and Wang

, TFRC-satellite: A TFRC variant with a loss differentiation algorithm for satellite networks, IEEE Transactions on Aerospace and Electronic Systems49(2) (2013), 716–725.

, Sun

, Wang

, et al., The effect of DTN custody transfer in deep-space communications, IEEE Wireless Communications20(5) (2013), 169–176.

Celandroni

, Ferro

, Gotta

, et al., On the applicability of reliable transport protocols in satellite delay tolerant and disruptive networks, International Journal of Satellite Communications and Networking32(2) (2014), 141–161.

10.

Gupta

and Goyal

, A review on congestion control in Ad-Hoc network[J], International Journal of Computer Science and Mobile Computing3(7) (2014), 28–33.

11.

Conti

and Giordano

, Mobile ad hoc networking: Milestones, challenges, and new research directions[J], Communications Magazine, IEEE52(1) (2014), 85–96.

12.

Mbarushimana

, Shahrabi

and Buggy

, A cross-layer TCP enhancement in QoS-aware mobile ad hoc networks[J], Computer Networks57(1) (2013), 286–301.

13.

Oliveira

, Furtadoa

, Bernardoa

, et al., Improving path duration in high mobility vehicular ad hoc networks[J], Ad Hoc Networks11(1) (2013), 89–103.

14.

Chowdhury

K.R.

, Di Felice

and Akyildiz

I.F.

, TCP CRAHN: A transport control protocol for cognitive radio ad hoc networks[J], Mobile Computing, IEEE Transactions on12(4) (2013), 790–803.

15.

Jehan

, Radhamani

and Kalakumari

, VEGAS: Better performance than other TCP congestion control algorithms on MANETs[J], International Journal of Computer Networks (IJCN)3(2) (2011), 151–158.

16.

Lochert

, Scheuermann

and Mauve

, A survey on congestion control for mobile ad hoc networks[J], Wireless Communications and Mobile Computing7(5) (2007), 655–676.

17.

Vijayalakshmy

, Lakshmy

and Sivaradje

, QoS improvement for the next generation heterogeneous network[J], International Journal of Computer Applications92(15) (2014), 32–44.

18.

, Lukyanenko

, Tarkoma

, et al., Tolerating path heterogeneity in multipath TCP with bounded receive buffers[J], Computer Networks64 (2014), 1–14.

19.

Shukla

R.S.

and Tyagi

, Global mibility and handover management for heterogeneous network in VANET [J], International Journal of Next-Generation Networks5(3) (2013), 23–34.

20.

, An improved congestion control strategy algorithm for heterogeneous network[J], Journal of Networks8(9) (2013), 2107–2113.

21.

Kim

H.A.

, Oh

and Lee

, Improvement of MPTCP Performance in heterogeneous network using packet scheduling mechanism[C]//Communications (APCC), 2012 18th Asia-Pacific Conference on IEEE, 2012, pp. 842–847.

22.

Yew

B.S.

, Ong

B.L.

and Ahmad

R.B.

, Performance evaluation of TCP vegas versus different TCP variants in homogeneous and heterogeneous networks by using network simulator 2[J], The Proceeding of International Journal of Electrical & Computer Sciences IJECS-IJENS11(03) (2011), 95–103.

23.

J.H.

, Feng

and Yeung

K.L.

, Hierarchical cache design for enhancing TCP over heterogeneous networks with wired and wireless links[J], Wireless Communications, IEEE Transactions on2(2) (2003), 205–217.

24.

Barakat

, Altman

and Dabbous

, On TCP performance in a heterogeneous network: A survey[J], Communications Magazine, IEEE38(1) (2000), 40–46.

25.

Nahm

, Helmy

and Kuo

C.C.J.

, Cross-layer interaction of TCP and ad hoc routing protocols in multihop IEEE 802.11 networks[J], Mobile Computing, IEEE Transactions on7(4) (2008), 458–469.

26.

Jain

, Padhye

, Padmanabhan

V.N.

, et al., Impact of interference on multi-hop wireless network performance[J], Wireless networks11(4) (2005), 471–487.

27.

Zhao

, Wang

, Wei

, et al., Model-based approach for available bandwidth prediction in multi-hop wireless networks[J], SCIENCE CHINA Information Sciences41(5) (2011), 592–604.

28.

Barakat

, Altman

and Dabbous

, On TCP performance in a heterogeneous network: A survey[J], Communications Magazine, IEEE38(1) (2000), 40–46.

29.

Manfredi

, Congestion control for differentiated healthcare service delivery in emerging heterogeneous wireless body area networks[J], Wireless Communications, IEEE21(2) (2014), 81–90.

30.

Dubois

, Fasson

, Donny

, et al., Enhancing TCP based communications in mobile satellite scenarios: TCP PEPs issues and solutions[C]//Advanced satellite multimedia systems conference (asma) and the 11th signal processing for space communications workshop (spsc), 2010 5th IEEE (2010), 476–483.

31.

Caini

and Firrincieli

, TCP Hybla: A TCP enhancement for heterogeneous networks[J], International Journal of Satellite Communications and Networking22(5) (2004), 547–566.

32.

Park

, Shin

, Oh

, Kim

and Lee

, TCP Hybla+: Making TCP more robust against packet loss in satellite networks, 2011 Computational Science and Its Applications-ICCSA, SpringerBerlin Heidelberg, 2011.

33.

C.P.

and Liew

S.C.

, TCP Veno: TCP enhancement for transmission over wireless access networks[J], Selected Areas in Communications, IEEE Journal on21(2) (2003), 216–228.