Available online at www.sciencedirect.com
ScienceDirect Procedia Computer Science 63 (2015) 529 – 532
The 2nd International Symposium on Intelligent Techniques in Distributed Systems (ITDS 2015)
A Server-based Flow Allocation Scheme for Streaming Services in Data Center Networks Yusuke Itoa , Yurino Satoa , Hiroyuki Kogaa , Katsuyoshi Iidab a Graduate b Global
School of Environmental Engineering, University of Kitakyushu, 1-1 Hibikino, Wakamatsu, Kitakyushu, 808-0135 Japan Scientific Information and Computing Center, Tokyo Institute of Technology, 2-12-1 Ookayama, Meguro, Tokyo, 152-8550 Japan
Abstract Streaming services provided by data centers (DCs) have become increasingly popular. The quality of streaming services greatly depends on network conditions, though, such as available bandwidth and delay time characteristics. To realize high quality streaming services, selecting an appropriate access network among available ones is necessary. However, in wireless LANs, a station (STA) simply selects an access point (AP) with the maximum received signal strength to connect to the Internet. Therefore, the service quality will deteriorate when access from STAs is concentrated in a specific AP. In this paper, we propose a server-based flow allocation scheme based on collectable information in DC networks to enable high quality streaming services. This scheme collects information, including the packet loss rate of existing flows as well as the available bandwidth of access networks, and then allocates a streaming flow to an appropriate access network based on the collected information. We show the effectiveness of this approach through simulation evaluations. c 2015 2015The TheAuthors. Authors.Published Published Elsevier © by by Elsevier B.V.B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of the Program Chairs. Peer-review under responsibility of the Program Chairs
Keywords: Flow allocation; Streaming service; Quality of service; Data center
1. Introduction Streaming services over the Internet is now one of the major services provided by data centers (DCs). Internet users can enjoy many forms of content, including movies and music, delivered by such services through widespread access networks such as wireless LANs located in coffee shops, commuter stations, and airports anytime, anywhere. The quality of streaming services greatly depends on network conditions such as available bandwidth and delay time characteristics 1 . To realize high quality streaming services, selecting an appropriate access network among the available ones is necessary. However, in wireless LANs which are worldwide access networks, a station (STA) simply selects an access point (AP) with the maximum received signal strength (RSS) to connect to the Internet. Therefore, the quality of services will deteriorate when access from STAs is concentrated in a specific AP 2 . ∗
Yusuke Ito. Tel.: +81-93-695-3254 ; fax: +81-93-695-3731 . E-mail address:
[email protected]
1877-0509 © 2015 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of the Program Chairs doi:10.1016/j.procs.2015.08.380
530
Yusuke Ito et al. / Procedia Computer Science 63 (2015) 529 – 532
,QIRUPDWLRQ
+DQGRYHUGLUHFWLYH
$YDLODEOHEDQGZLGWK 3DFNHWORVVUDWH
&OLHQW
$3 $ 6HUYHU
67$
6HUYHU
1HWZRUN
&OLHQW
$3 %
1HWZRUN
$3 'DWD&HQWHU
Fig. 1. Proposed scheme
7&0
,QIRUPDWLRQ $3 &
,QIRUPDWLRQFROOHFWLRQ )ORZDOORFDWLRQ
6HUYHU
6HUYHU
$3 $FFHVV3RLQW 7&0 7UDQVSRUW&RQQHFWLRQ0DQDJHU
Fig. 2. Unified Central Congestion Control Architecture
In this context, an AP selection strategy has been proposed to enable efficient utilization of wireless network resources 3 . This approach autonomously selects an AP to be connected by STAs based on the available bandwidth of wireless networks. Each STA collects available bandwidth information from connectable APs and then selects an appropriate AP based on the collected information. This approach enables effective use of wireless network resources among STAs, but the quality of services will be degraded when congestion occurs in core networks even if wireless access networks provide good conditions. Therefore, to achieve a high quality service, an appropriate AP must be selected based on network conditions in core networks in addition to those in access networks. In this paper, we propose a server-based flow allocation scheme based on collectable information in DC networks to achieve high quality streaming services. This scheme collects information, including the packet loss rate of existing flows as well as the available bandwidth of access networks, and then allocates a streaming flow to an appropriate access network based on the collected information. We have demonstrated the effectiveness of this approach through simulation evaluations.
2. Proposed Scheme This scheme allocates a streaming flow to an appropriate access network based on collectable information in DC networks as shown in Fig. 1. The information is collected by Unified Central Congestion Control Architecture (UC3 ) 4 which uniformly manages congestion information in networks as shown in Fig. 2. A Transport Connection Manager (TCM) located in DC networks collects congestion information such as the number of existing flows, flow characteristics, and the available bandwidth from routers and servers in DC networks as well as APs in access networks, and then notifies the servers of appropriate flow settings satisfying the quality of service requirements from applications based on the collected information to provide flow allocation to an appropriate path, server-controlled handover, and so on. In this scheme, two kinds of information are collected to perform flow allocation. One is the available bandwidth of access networks. The servers are periodically notified of the available bandwidth information from APs through TCM in the DC networks so that they can determine an appropriate access network satisfying a required rate from a client. The APs add the available bandwidth information in the header of ACK packets sent from STAs to the servers located in the same DC networks to report it. The other is the packet loss rate of existing flows. This is collected from the servers through TCM in the DC networks to select an appropriate access network with good conditions for a transmission path from the servers to the STAs. The servers measure it from existing flows toward the same access networks. When an STA demands a streaming service, it sends a request packet with connectable AP information to a server. The server receiving the request packet selects an appropriate AP with the lowest packet loss rate among the connectable APs with an available bandwidth satisfying the required rate. If the STA is now connecting to a different AP, the server informs it of the appropriate AP information. The STA receiving the appropriate AP information reconnects
531
Yusuke Ito et al. / Procedia Computer Science 63 (2015) 529 – 532 6HUYHU 5RXWHU $FFHVV3RLQW &OLHQW :LUHG :LUHOHVV
6 0EV
PV 5
0EV
0EV
PV
0EV
$3
$3
0EV
PV
PV $3
&K
$3
5 0EV
$3
&
$3
PV
5 0EV PV
67$
&
PV
$3
&L
&
$3
&M
&
&N
,((( J
Fig. 3. Simulation topology
Fig. 4. Location of STAs
to the notified AP and again sends a request packet. Through these procedures, the proposed scheme achieves a high quality streaming service through server-based flow allocation taking into consideration the network conditions in both access and core networks.
3. Simulation Model To investigate the effectiveness of our scheme, we evaluated it through simulations using Network Simulator ns-3 5 after implementing the proposed scheme. Figure 3 shows the simulation topology. In this simulation, 25 STAs are communicating with the server through 4 APs at the beginning of the simulation. To bias the load of APs, each STA is located and connected to a nearby AP as shown in Fig. 4. Namely, many STAs connect to AP1. We assume that each AP is operated by a different service provider — i.e., there are different access networks — providing a data rate of 54 Mb/s (per IEEE 802.11g) with different radio channels to prevent interference among the APs. Five seconds after the simulation starts, additional STAs are connected to the APs one by one at intervals of 1 second. The total number of STAs is 33. All STAs can connect to all APs. Each STA requests a streaming service with the same data rate, which ranges from 0.8 to 1.8 Mb/s. The transport layer protocol employed is NewReno TCP. In this simulation, we evaluated the total throughput performance of the proposed scheme as compared with that of conventional schemes which select an AP based on either RSS or the available bandwidth. The first type of conventional scheme is defined as a conventional (signal) scheme and the second type is defined as a conventional (bandwidth) scheme. In the conventional (signal) scheme, the additional STAs are certainly connected to AP1, while they are connected to APs according to selection algorithms in the other schemes.
4. Simulation Results Here, we show simulation results and discuss the effectiveness of the proposed scheme as compared with the conventional schemes. Figure 5 shows the total throughput performance of the proposed and conventional schemes when the number of additional STAs varies from 0 to 8. The required rate of streaming service is set to 1 Mb/s. When the number of additional STAs is small (from 1 to 3), there is no difference in the total throughput between the proposed and conventional (bandwidth) schemes because these schemes select AP2 since there is no congestion in the core networks. On the other hand, when the number of additional STAs is larger than 3, the proposed scheme achieves higher throughput than the conventional schemes. In this case, although congestion occurs on the transmission paths through AP1 or AP2 in the core networks, the conventional scheme (bandwidth) selects AP2 because its available
32 30 28 26 24 22 20 18
Total throughput [Mb/s]
Yusuke Ito et al. / Procedia Computer Science 63 (2015) 529 – 532
Total throughput [Mb/s]
532
Proposed Conventional (Bandwidth) Conventional (Signal)
0
1
2 3 4 5 6 Number of additional STAs
Fig. 5. Effect of number of additional STAs
7
8
28 26 24 22 20 18 16 14
Proposed Conventional (Bandwidth) Conventional (Signal) 0.8
1
1.2 1.4 Required rate [Mb/s]
1.6
1.8
Fig. 6. Effect of required rate
bandwidth is adequate. However, the proposed scheme selects AP3 where congestion does not occur in core networks by effectively using the packet loss rate information. Next, Fig. 6 shows the total throughput performance of the proposed and conventional schemes when the required rate varies from 0.8 to 1.8 Mb/s. Here, the number of additional STAs is set to 5. As shown, the proposed scheme enables higher throughput than the conventional schemes regardless of the required rate. In particular, when the required rate is set from 1.0 to 1.6 Mb/s, the difference in the total throughput between the proposed and conventional schemes is large. This is because the proposed scheme can effectively utilize an available bandwidth by selecting the appropriate transmission path where congestion does not occur in the core networks. 5. Conclusion In this paper, we have proposed a server-based flow allocation scheme based on collectable information in DC networks to enable high quality streaming services. Simulation evaluations have indicated that the proposed scheme provides higher throughput than conventional schemes by selecting an appropriate AP based on network conditions in both access and core networks. In our future work, we will consider effective ways to apply this approach in a real environment, such as where streaming flows with different required rates coexist. 6. Acknowledgements This work was supported in part by the Japan Society for the Promotion of Science, Grant-in-Aid for Scientific Research (B) Number 25280028. References 1. T. Huang, R. Johari, N. McKeown, M. Trunnell, and M. Watson, “A buffer-based approach to rate adaptation: Evidence from a large video streaming service,” ACM SIGCOMM Computer Communication Review, vol. 44, no. 4, pp. 187–198, 2014. 2. Y. Bejerano, S. Han, and L. Li, “Fairness and load balancing in wireless LANs using association control,” IEEE/ACM Transactions on Networking, vol. 15, no. 3, pp. 560–573, 2007. 3. Y. Fukuda and Y. Oie, “Decentralized access point selection architecture for wireless LAN,” IEICE Transactions on Communications, vol. E90–B, no. 9, pp. 2513–2523, 2007. 4. K. Kusuhata, I. Kaneko, K. Iida, H. Koga, and M. Shimamura, “A unified congestion control architecture design to improve heterogeneous wireless network efficiency and accommodate traffic by various rich applications,” Proc. IFIP HET-NETs2013, 5 pages, Nov. 2013. 5. The ns-3 network simulator, http://www.nsnam.org/.
