Using 802.11 MAC Retransmissions For Path Selection in Multi-homed Transport Layer Protocols Sheila Fallon, Paul Jacob, Yuansong Qiao, Austin Hanley
Liam Murphy
Software Research Centre, Athlone Institute of Technology, Ireland
[email protected], {pjacob,ysqiao,ahanley}@ait.ie Abstract— The Stream Control Transmission Protocol (SCTP) is a transport layer protocol which can support mobility through its multi-homing feature. One of the key parameters used by SCTP to manage path selection is Retransmission Time Out (RTO). We illustrate that SCTP’s evaluation of underlying paths using end to end metrics, results in the calculation of excessive RTO for degraded wireless paths. This excessive RTO causes SCTP to behave in a counter intuitive manner which delays path switchover. Our investigation shows that the increased Round Trip Time (RTT) is as a result of retransmissions at the 802.11 MAC. We illustrate that when a connectionless 802.2 LLC service is utilized, 802.11 independently implements positive frame acknowledgement resulting in significantly increased RTT. We propose a cross layer path selection algorithm, which recognizes increased 802.11 MAC retransmissions as an indicator of imminent path failure. Results presented indicate that our cross layer approach has excellent performance in comparison to standard SCTP strategies. Keywords- SCTP, RTO, MAC Retransmissions, Mobility
I.
INTRODUCTION
The trend towards heterogeneous networking has led to a change in focus for network planners. Originally wireless networks provided coverage in specific “hot spot” zones. With the emerging significance of real time applications such as Voice Over IP (VOIP), software applications are required to respond actively to changing network conditions. Using mobility enabled protocols such as SCTP, applications can seamlessly migrate between wireless networks. The dynamic characteristics of wireless networks can result in significant variations in performance in a short period of time. Mobility protocols must react dynamically to these performance variations. SCTP [1][2][3] provides support for transparent mobility through multi-homing - the ability to implement an end-to-end communication session transparently over multiple physical paths where each end-point is identified by an IP address. In this paper we propose that the path selection functionality in SCTP be replaced by a cross layer algorithm which uses 802.11 MAC retransmissions as an indicator of path performance. In earlier work [4][5] we presented results showing that excessive RTO for wireless paths caused SCTP to behave in a counterintuitive manner by significantly delaying path failover.
School of Computer Science & Informatics University College Dublin, Ireland
[email protected] While in [6], we illustrated that the excessive RTO results from continually increased access network RTT caused by the division of functionality between 802.2 Logical Link Control (LLC) and the 802.11 Medium Access Control (MAC). The 802.11 MAC retransmits packets regardless of the service type requested by upper layers [7]. As a transport layer protocol SCTP adheres to rigid layer boundaries. When an SCTP packet is transmitted on a degrading wireless path, the 802.11 MAC may retransmit the packet many times before successful delivery. The repeated retransmission of packets by the 802.11 MAC delays the successful transmission of the SCTP packet from the mobile node to the access point. These repeated retransmissions also delay the return SCTP ACK. through the access point to the mobile node. It is only when the ACK is received that SCTP calculates RTT and begins its performance evaluation of the underlying path. While SCTP’s reactive performance evaluation is a limitation even for a functioning environment, in a degraded wireless environment this reactive approach significantly degrades performance. We propose a cross layer algorithm which uses 802.11 MAC retransmissions as an indicator of performance for all paths within an association. Previous studies [8] have indicated that in a WLAN environment there exists a point of performance transition from a low loss, low delay regime to a high-loss, high delay operation. This sudden transition in performance is observed as a mobile node moves from the zone of coverage of the AP hosting the primary path. By defining a cross layer approach which utilizes 802.11 MAC retransmissions, we can accurately predict this performance transition significantly earlier than at the transport layer. Results presented illustrate that such an approach is an excellent indicator of path performance. We present a scenario in which our cross layer algorithm successfully transmits approximately twice the data of the standard SCTP based approaches. II.
RELATED WORK
Previously [4][5][6] we identified that SCTP path switchover functionality, which uses end to end performance metrics, caused delayed switchover and reduced throughput in wireless environments. In [4] we illustrated that increased RTT significantly distorted RTO calculations causing SCTP to behave in a counterintuitive manner which allowed more time for switchover as network conditions degraded. In [5] we
978-1-4244-4148-8/09/$25.00 ©2009 This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2009 proceedings.
evaluated whether a transport layer oriented approach, achieved through the alteration of standard SCTP RTO parameters α, the smoothing factor, and β, the delay variance factor could address the performance limitations highlighted in [4]. It was shown that while performance improvements were achievable, the switchover delays highlighted could not be addressed through a transport layer specific solution. In [6] we proposed to address the performance limitations in [4] using a cross layer approach which uses network layer, RTT between the mobile node and AP, as well as transport layer end to end performance characteristics for path performance evaluation. Such an approach was shown to have up to a 35% performance improvement over the suggested standard SCTP strategy. In this paper we propose a link layer, transport layer approach to path selection. We propose that the number of MAC layer packet retransmissions be used as a performance metric to manage transport layer path migration. We illustrate MAC retransmissions are an excellent indicator of path performance and result in significantly increased throughput. A number of studies have identified performance implications regarding the 802.11 retransmission mechanism. In [9] an explicit retransmission notification is introduced for use by upper layer congestion control procedures. In [10] it is indicated that local retransmission between the 802.11 MAC and access points distort RTT calculation for TCP. A mechanism is introduced which hides the duration of the 802.11 MAC error recovery phase from upper layer protocols. In [11] it is proposed that link-layer retransmissions result in delays which cause undesired control actions in TCP. Additional delays are inserted in link layer packets which results in the calculation of more appropriate TCP values. These studies are for 802.11 delay spikes for TCP and do not consider SCTP or mobility with degraded Receiver Signal Strength (RSS) for wireless networks. In [8] the tradeoff between access point buffering and loss for voice traffic in 802.11 networks is investigated. As a result of mobility an abrupt transition from the low-loss, low delay regime to high-loss, high-delay operation is observed. The study indicates that below the point of performance transition, AP buffer size has little impact on throughput. However as the point of performance transition is passed total delay depends strongly on AP buffer size. In this paper we observe this point of performance transition. We suggest however, that 802.11 MAC retransmissions rather than the buffer sizes are the primary cause of increased RTT. A large number of studies have been undertaken which optimize SCTP path migration. In [12] it was suggested that aggressive failover strategies achieved through low Path.Maximum.Retransmissions (PMR) values did not degrade performance as spurious failovers, those caused by packet loss rather than path degradation, were temporary in nature and resulted in an immediate switchback. Furthermore, aggressive switching strategies were shown to improve goodput regardless of the alternate paths’ characteristics (bandwidth, delay, and loss rate). The aggressive failover approach in which PMR=0 has been accepted as the standard when implementing network mobility. We compare our cross layer path migration algorithm against this approach and illustrate that significant throughput increases are achievable.
In [13] an SCTP mobility management scheme is introduced which uses link layer RSS to manage path selection. The approach is shown to improve performance over Mobile IP (MIP). Two paths with equivalent RSS can have different performance characteristics resulting from e.g. load differences. Our algorithm has advantage over an RSS based approach of differentiating the throughput capability of all paths within an association. In [14] an SCTP handover scheme for VoIP applications for heterogeneous networks is presented. The proposed handover scheme is based on the ITU-T E-Model for voice quality. EModel Mean Opinion Score (MOS) metrics are calculated for all paths within an association. The path with the highest MOS is selected as primary. While in [15] it is suggested that the SCTP handover strategy is reactive in nature and a more proactive approach where handover is based on path delays should be introduced in order to pre-empt and avoid path failures. We suggest that the delayed delivery of packets to upper layers caused by 802.11 MAC retransmissions will limit the accuracy of these approaches. III.
IMPACT OF 802.11 MAC RETRANSMISSIONS ON SCTP PERFORMANCE
Through 802.2 LLC the IEEE have defined a uniform interface by which upper layers can access the services of carrier protocols. Below the LLC a MAC sublayer implements packet transmission for specific medium types such as 802.3, 802.11 and 802.16. There are 3 types of service specified by 802.2; Type 1 is an unacknowledged connectionless service and does not guarantee in order delivery of packets, Type 2 is connection-oriented and guarantees in order delivery using sequence numbers while Type 3 is an acknowledged connectionless service. In order to illustrate how MAC layer packet retransmissions affected RTT an experimental test configuration was created which consisted of two Laptops, Laptop 1 representing a mobile client and Laptop 2 a back end server connected by a Linksys WRT54GL 802.11g access point. Both laptops were configured with SCTPLIB [16]. The test started adjacent to the access point. The mobile node then moved at slow walking pace a distance of 100M from the access point. Our experimental configuration used LLC Type 1 service, as verified by the Type 1 indicator 0x03 in the traced packets. In particular we note that while 802.11 uses an 802.2 Type 1 connectionless service, it implements its own positive frame acknowledgment independently of 802.2 through an 802.11 ack control frame.[7] In this way the 802.11 MAC at the receiving station implements error control independently of 802.2. If the sending station doesn't receive a MAC level ACK after a period of time, it will retransmit the frame. The manner by which SCTP RTO is calculated is critical to this investigation. The deficiencies in SCTP RTO calculation are best illustrated using a single test. As the mobile node moves from the coverage area of the access point signal strength degrades and results in intermittent network connectivity. As the signal strength degrades the RTT and
978-1-4244-4148-8/09/$25.00 ©2009 This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2009 proceedings.
corresponding RTO increase. Figure 1 illustrates the RTO and SRTT for the selected test. 50000
30000 20000 10000 0 0
SRTT (ms)
40000 RTO (ms)
2000 1800 1600 RTO 1400 1200 1000 800 600 400 200 0 10 20 30 40 50 60 70 80 90 100 110 120 130 SRTT
Test Time (Secs)
Figure 1. Increased SRTT and RTO
Number of MAC Retransmissions
Figure 2 details the number of 802.11 MAC retransmissions, for the period 0-30 seconds. It illustrates that as RSS degrades the number of 802.11 MAC retransmissions increases significantly. 350
At 33.398 secs an SCTP packet with TSN 38997 was transmitted by the 802.11 MAC. This packet required 3 retransmissions before it was successfully delivered. The table also shows that similar MAC retransmissions were required to deliver a SACK for packets 11893 and 38933. Retransmissions of 802.11 packets occur until 34.439secs when the 802.11 MAC is finally available to SACK SCTP TSN 38997 resulting in an RTT of 1.04sec for SCTP packet TSN 38997. The preservation of transport layer protocol boundaries significantly impacts SCTPs ability to extract meaningful path performance metrics. It is only when a packet has been received at the transport layer that SCTP can begin to analyse path performance. In the situation illustrated in Table 1 this performance evaluation only begins 1.04 secs after the packet is first transmitted by the MAC layer. The action SCTP takes is inappropriate as it assumes that the increased RTT results from Internet congestion. The increase in RTO significantly delays the detection of loss packets and therefore delays switchover from a degraded path. IV.
300 250 200 150 100 50 0 0
5
10
15
20
25
30
Test Time (Secs)
Figure 2. Number of 802.11 MAC Retransmissions
In order to illustrate how MAC retransmissions increase SCTP RTT we analyse a Wireshark [17] trace using AIRPCAPs 802.11 Wireless Packet capture plug-in at approximately 33 seconds. A selection of packet transmissions for this period are detailed in Table I. TABLE I.
AIRPCAP WIRESHARK TRACE WITH MAC RETRANSMISSIONS
Time
Source
Destination
Protocol Type
Information
33.398 33.399 33.40 33.401 33.402 33.403 33.404 33.406 33.415 33.417 33.418 33.421 . 34.439
192.168.1.100 192.168.1.100 192.168.1.100 192.168.1.100 192.168.1.100 192.168.1.100 192.168.1.100 192.168.1.100 192.168.1.110 192.168.1.110 192.168.1.110 192.168.1.110 . 192.168.1.110
192.168.1.110 192.168.1.110 192.168.1.110 192.168.1.110 192.168.1.110 192.168.1.110 192.168.1.110 192.168.1.110 192.168.1.100 192.168.1.100 192.168.1.100 192.168.1.100 . 192.168.1.100
SCTP SCTP SCTP SCTP SCTP SCTP SCTP SCTP SCTP SCTP SCTP SCTP . SCTP
DATA (TSN = 38997) DATA (TSN = 38997) DATA (TSN = 38997) DATA (TSN = 38997) SACK TSN 11983 SACK TSN 11983 SACK TSN 11983 SACK TSN 11983 SACK TSN 38933 SACK TSN 38933 SACK TSN 38933 SACK TSN 38933 . SACK TSN 39087
A CROSS LAYER PATH MIGRATION ALGORITHM
The rigid transport layer boundary within which SCTP operates, dictates that path performance evaluation can only occur following packet reception at the transport layer. However, degrading path performance results in increased RTT thereby creating a time lapse between path degradation at the physical layer and its detection at the transport layer. Results presented in Figure 1 and Figure 2 illustrate that MAC retransmissions are an excellent early indicator of path performance. In Figure 1 the number of MAC retransmissions increase from 25 seconds while in Figure 2 the subsequent RTT begins to increase from approximately 30 seconds. Therefore we propose a cross layer path selection algorithm which uses MAC retransmissions as an indicator of path performance and SCTP to implement switchover. A. MAC Layer Performance Parameter Selection Our algorithm will use the number of MAC retransmissions as input. Since increased RTT and increased MAC retransmissions have similar performance trends, we use smoothing similar to that used for RTO calculation [1][18] to address wireless delay spikes for our cross layer path selection algorithm. Our algorithm takes the number of 802.11 MAC retransmissions as input. In order to define normal retransmission behavior a SmoothRetrans variable is introduced which is calculated as follows: SmoothRetrans.new = (1- α) SmoothRetrans.old+ α RTrans.new SmoothRetrans is the value against which retransmission counts are compared to determine normal behavior. In general if the number of retransmissions is above the SmoothRetrans then the path is in a degrading condition. In the same way, if the number of retransmissions is below SmoothRetrans the condition of the path is improving. Initial investigations we undertook also utilized a RetransVar parameter similar to that
978-1-4244-4148-8/09/$25.00 ©2009 This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2009 proceedings.
Figure 3. Effect of α alterations on Retransmissions Smoothing
For the period from approximately 40 to 80 secs the path was in a degrading condition. For the periods 60 to 65 secs and 75 to 80 secs the number of retransmissions was less than SmoothRetrans for α = .1 resulting in false positive diagnosis of the path. For α = .025 the value of SmoothRetrans and the number of retransmissions are similar at 62 secs, resulting in a false diagnosis of the path. Figure 3 illustrates that .01 is a suitable choice for α as it results in SmoothRetrans values which are responsive to degrading network conditions while also minimizing the possibility of false positive diagnosis. Figure 4 also indicates that α = .01 is an appropriate value when path conditions are improving. B. Defining Normality Boundaries and Performance Indicators In order to determine path performance we define scaled normality boundaries as in Table II and Table III. The scaled normality boundaries define the extent to which retransmission values deviate from normal behavior defined by SmoothRetrans. We associate a path performance indicator with each normality boundary from 0, excellent performance indicating mobile node is close to access point to 5, poor performance indicating that the mobile node is moving away from the access point. TABLE II.
NORMALITY BOUNDARIES FOR DEGRADING WIRELESS PATHS
Normality Boundaries Retrans < SmoothRetrans*.1.5 SmoothRetrans*.1.5
