單元 7 LTE 與 WiFi 的負載分享排程 演算法 ( Load Scheduling)

1. 教育部行動寬頻尖端技術人才培育計畫-小細胞基站聯盟中心教育部行動寬頻尖端技術人才培育計畫-小細胞基站聯盟中心 「小基站與WiFi之異質性網路存取」課程模組 單元7LTE與WiFi的負載分享排程演算法(Load Scheduling) 助理教授：吳俊興 助教：王瑞元 國立高雄大學 資訊工程學系

2. Outline • Overview • LWIP and Wi-Fi Boost Flow Control • LWIR: LTE-WLAN Integration at RLC Layer with Virtual WLAN Scheduler for Efficient Aggregation • A Fair Listen-Before-Talk Algorithm for Coexistence of LTE-U and WLAN • 在OAI平台的LWA網路上使用UDP隧道實作影音串流的雙介面傳送機制 • Summary

3. Overview • LTE macro-cells can provide high speed mobile data • But they can’t fulfill the increased subscribers’ demands in high density places • Heterogeneous networks (HetNet) incorporate the coexistence of low power nodes along with a macro base station to improve coverage in high demand areas • Seem to be a promise low cost solution from operators' point of view • The HetNet variations are categorized into a single RAT (Radio Access Technology), multi-tier network, and a multi RAT multi-tier network • The possible single RAT scenarios are : micro-cell, pico-cell, femto-cell or fixed relays (small cell) • For the multi RAT scenario, the possible network components are: Wi-Fi offload, mobile hotspots and virtual carrier

4. ThreeMain Levels of Integration between Wi-Fi Networks andCellular Networks

5. Conventional Wi-Fi Offloading Algorithms • Network access selection or offloading decisions, are made based on different parameters with different objectives as well, such as boosting the capacity or user QoS • Two conventional algorithms • Wi-Fi First algorithm (WF): Wi-Fi if coverage, which considers the SNR value to trigger the offloading decision to Wi-Fi • Fixed SNR Threshold: based on choosing the best SNRmin for WLAN APs

6. Enhanced Offloading Algorithms • Best- Server algorithm • Happens if SINR perceived by the user is greater than a certain threshold; otherwise he will connect to an LTE network • Physical Data Rate Based algorithm (PDR) • purely based on the PDR provided by different available RATs • Compare the PDR for Wi-Fi and LTE, and chooses the highest value a user can get • SMART algorithm • Trigger the offloading based on the minimum data rate perceived by the user as long his experienced SINR exceeds the threshold

7. Effective CapacityWi-Fi Offloading Algorithms

8. Case Studies • LWIP and Wi-Fi Boost Flow Control • A new feature of Wi-Fi Boost, its radio link management, which allows to smartly steer the downlink traffic between both LTE and Wi-Fi upon congestion detection • LWIR: LTE-WLAN Integration at RLC Layer with Virtual WLAN Scheduler for Efficient Aggregation • Virtual WLAN Scheduler (VWS) which employs traffic steering scheme • Five different bearer selection schemes have also been proposed which provide efficient steering by smartly choosing a bearer to route some data onto Wi-Fi

9. Case Studies (Cont.) • A Fair Listen-Before-Talk Algorithm for Coexistence of LTE-U and WLAN • F-LBT jointly considers the total system throughput and the fairness between LTE-U and WLAN, and then allocates an appropriate idle period for WLAN • 在OAI平台的LWA網路上使用UDP隧道實作影音串流的雙介面傳送機制 (中山大學) • 實作兩種雙介面傳送的技術，並且使用影音串流的資料來傳輸，其中第一種是以串流為基礎，我們是將整條串流經由 LTE 介面或 Wi-Fi 介面傳送，第二種是以支流為基礎，我們是將一條串流分成 LTE 及 Wi-Fi 兩條支流，並且將兩條支流分別經由 LTE 及 Wi-Fi 介面傳送

10. Outline • Overview • LWIP and Wi-Fi Boost Flow Control • LWIR: LTE-WLAN Integration at RLC Layer with Virtual WLAN Scheduler for Efficient Aggregation • A Fair Listen-Before-Talk Algorithm for Coexistence of LTE-U and WLAN • 在OAI平台的LWA網路上使用UDP隧道實作影音串流的雙介面傳送機制 • Summary • References

11. LWIP and Wi-Fi Boost Flow Control • Introduction and Background • Data explosion • 5G • New Radio Access Technologies : Wi-Fi Aggregation Solutions • LWIP(LTE WLAN Radio Level Integration with IPsec Tunnel)) • Wi-Fi Boost : Advantages • Architecture • Compare with LWIP • Algorithm • Simulation & Result • Conclusion

12. Introduction and Background • The Mobile Network in 2016 • Global mobile data traffic grew 63 percent • Mobile data traffic has grown 18-fold over the past 5 years • Fourth-generation (4G) traffic accounted for 69% of mobile traffic • Mobile offload exceeded cellular traffic by a significant margin • Sixty percent of total mobile data traffic was offloaded onto the fixed network through Wi-Fi or femtocell • 10.7 exabytes each month (1 EB = 1 billion GB) • Almost half a billion (429 million) mobile devices and connections were added

13. CISCO Forecast Source :https://www.cisco.com/c/en/us/solutions/collateral/service-provider/visual-networking-index-vni/mobile-white-paper-c11-520862.html

14. 5G Standardization • Enhanced Mobile Broadband • Massive Machine Type Communications • Ultra-Reliable Low-Latency Communications Source : 3GPP TR 22.891 v14.1.0 (2016-06)

15. New Radio Access Technologies Wi-Fi Aggregation Solutions Source : https://www.netmanias.com/en/?m=view&id=reports&no=8532

16. LWIP(LTE WLAN Radio Level Integration with IPsec Tunnel) • NetworkLayer, Internet protocol • Encrypt the original packet • Add a new set of IP addresses • The connection between the eNB and the UE is transmitted by the IPsec tunnel • LWIP-SeGW Source : 3GPP TS 36.300 v14.4.0 (2017-09)

17. Wi-Fi Boost : Advantages • Wi-Fi Boost uses LTE access for UL and frees up the enterprise’s existing Wi-Fi network for DL • The solution works without any hardware or software upgrade on Wi-Fi infrastructure, and only requires a software upgrade on LTE eNBs and UEs • Local access allows the UE to choose either LTE or Wi-Fi for UL applications anchored in the enterprise core

18. Architectures Commonalities • Both Wi-Fi Boost and LWIP R13 use as DL anchor the LTE eNB, and utilize IP layer as the split/aggregation point • Both technologies are able to take advantage of the DL and UL split concept i.e. UL on LTE and DL on Wi-Fi • An IPsec tunnel is used to transmit DL traffic from the LTE eNB to the UE through the Wi-Fi AP in a secure manner

19. Architectures Differences • Wi-Fi Boost uses over the top proprietary signaling to establish the IPsec tunnel, which only requires a software upgrade on LTE eNBs • The Wi-Fi Boost solution allows for a more powerful radio link management, at the expense of the software upgrades required at the UE side in order to realize the necessary cooperation/feedback • IP re-ordering and duplicate discard at the UE is another distinctive feature that can be made available in Wi-Fi Boost

20. Architectures

21. Wi-Fi Boost compare with LWIP

22. Radio Link Management – Initial Phase • Upon connection request, the LTE eNB will estimate which is the most suitable path for the given UE • RAN connection manager (RCM) • i) Fraction of probes lost, • ii) Average probe delay, • iii) Average probe throughput, • The fraction of probes lost is lower than a threshold, < • The average probe delay is shorter than a threshold, < • The average probe rate is higher than a threshold, >

23. Radio Link Management – Data Phase • Stall detection • If consecutive active probe ACKs are missing • The RCM switches the UE over the LTE path • Inactivity detection • If the number of bits transmitted in between two probe ACKs is smaller than a threshold, , meaning that the UE generates a small amount of traffic, the RCM switches the UE over the LTE path and the UE may decide to switch to Wi-Fi only mode to save resources • Congestion detection • If the filtered average UE throughput is smaller than a threshold, , the RCM switches the UE over the LTE path

24. Flow Control Algorithm

25. Simulation&Result – Throughput CDF • The UE throughput cumulative distributionfunction (CDF) for the case where there are 4 UEs in the enterprise • The LTE only case provides a median throughput of 21.93 Mbps/UE, while the Wi-Fi only case provides a larger median throughput of 60.93 Mbps/UE • The WiFi only case benefits from more cells • LWIP R13 and Wi-Fi Boost have a substantial gain over the Wi-Fi only case of around 2x • The offloading of UL traffic from the unlicensed to the licensed band and theresulting collision-free usage of the unlicensed spectrum for DL (the so-called Boost effect) • Comparison to that of LTE due to the inefficient sharing of resources between nodes and the contention/collision issues in the former

26. UE Throughput CDF

27. Simulation&Result – Throughput Distribution • The UE throughput distribution for the case where there are 32 UEs in the enterprise • The even larger traffic load, and the resulting larger contention and congestion • The gap between the performance of the LTE only and Wi-Fi only cases reduces further • How CSMA/CA becomes more and more inefficient as the traffic load increases • The larger congestion, the performance gain of LWIP R13 and Wi-Fi Boost with respect to the Wi-Fi only case is again larger • LWIP R13 and WiFi Boost can enhance network performance up to 5x and 6x over LTE only, and 4x and 5x over Wi-Fi only networks • Fig. 6 shows the system throughput (sum throughput of • all cells) distribution for the case where there are 32 UEs in the enterprise

28. UE and System Throughputs

29. Conclusion • Simulation performance • Up to 5x and 6x over LTE-only • 4x and 5x over Wi-Fi only networks • Further enhance over LWIP R13 up to 19 % • Machine learning techniques • Optimize the algorithm parameters • Suggests to provide UE estimations to the LTE eNB on short-term throughput to detect congestion • Enhance LWIP R13 UE feedback in future LTE releases

30. Outline • Overview • LWIP and Wi-Fi Boost Flow Control • LWIR: LTE-WLAN Integration at RLC Layer with Virtual WLAN Scheduler for Efficient Aggregation • A Fair Listen-Before-Talk Algorithm for Coexistence of LTE-U and WLAN • 在OAI平台的LWA網路上使用UDP隧道實作影音串流的雙介面傳送機制 • Summary • References

31. LWIR: LTE-WLAN Integration at RLC Layer • Backgroundand Motivation • 5G Standard Feature • Unlicensed band • 3GPP LTE-H (Heterogeneous) technology • Proposed Work • LWIR Architecture • VWS : One-Sized windowing technique • VWS : Bearer/User Selection Schemes • VWS : Feedback mechanism • Simulation Setup and Performance Result • Conclusion and Future Work

32. Backgroundand Motivation • Cisco Visual Networking Index: Global Mobile Data Traffic Forecast Update, 2016–2021 • Global mobile data traffic grew 63 percent in 2016 • Mobile data traffic has grown 18-fold over the past 5 years • Fourth-generation (4G) traffic accounted for 69% of mobile traffic in 2016 Global Mobile Devices and Connections Growth Global Mobile Traffic Growth by Device Type

33. 5GStandard Feature • High data rate and capacity • Low-Latency and Ultra-Reliable • High energy efficiency • Support IoT device • New Radio Access Technology • Massive MIMO • Use more Spectrum : 10GHz ~ 100GHz • Multi-cell Connectivity • Wider bandwidth per carrier

34. Unlicensed Band • The wireless spectrum is limited • Increase the capacity of the cellular system • LTE-U : 5GHz • LTE-H : 2.4GHz, 5GHz(Wi-Fi) • LWA, LWIP

35. 3GPP LTE-H Technology • LWA, LWIP • LWIR • Reduces out-of-order delivery at the UE side • When to switch between the radios • How much data to route/steer through Wi-Fi

36. Variation in Congestion Window Size in LTE and LWA • Traffic control is mainly by size of Sliding Window (window size) to adjust • If the receiver is too busy to handle the packet, the windows will become smaller

37. Proposed Work • LWIR Architecture • Traffic Steering Granularity • How much data to steer? • Bearer Selection for Wi-Fi Offloading • Virtual WLAN Scheduler (VWS) • One-sized windowing technique • Bearer/User Selection Schemes • Feedback mechanism

38. One-sized Windowing Technique • Keeps a limit on the amount of data offloaded to Wi-Fi • At most one packet kept in the Wi-Fi queue • Never suffers from WiFi MAC queuing delay • The VWS limits the size of RLC SDU to Maximum Transmission Unit (MTU) when it fetches from RLC buffer • As LTE is a scheduled interface • The achievable rate is deterministic • Wi-Fi with random access based transmission gets increased throughput • Increasing payload size of the packet • The packet size (MTU) increases • The number of transmitted packets decreases • Reduces the total channel access delay

39. Bearer/User Selection Schemes • VWS picks the data by selecting a RLC buffer • Based on the Bearer/User selection scheme • Channel Quality Indicator (CQI) • Min CQI First • Interference region or at the edge of the cell • Max CQI First • Nearby the LWIR node receive better signal strength from both the radios • Max RLC Buffer First • Having highest amount of data in its RLC buffer • Ensures maximum steering onto Wi-Fi • Max RLC Buffer with Min CQI • Sufficient amount of data in their RLC buffers • Other user who has the least CQI is selected for steering onto Wi-Fi • Max RLC Buffer with Max CQI

40. Feedback Mechanism • Fair resource allocation • Conveys amount of data being steered onto the Wi-Fi network to the LTE scheduler • In Proportional Fair scheduler, the user selection priority function is given by

41. Simulation Setup • LWIR architecture along with VWSfunctionalities in NS-3 • LWIR node uses Wi-Fi Only in Downlink • One Macro eNB and one LWA/LWIR/LWIP node in the simulation scenario • 30 UEs :each UE one downlink flow • Tested with TCP based flows

42. Performance Results • Total throughputs of TCP flows for flow level traffic steering in LWIP and split bearer traffic steering with varying split ratios at packet granularity in LWA • Flow level traffic steering in LWIP which offloads some flows completely onto Wi-Fi • Flow level traffic steering has rare chance of out-of-order packet reception at UE • High number of DUPACKS • With increase in the value of PDCP packet re-ordering timer, number of DUPACKS decreases • The main cause for less throughput in LWA is waiting delay in Wi-Fi queue • Improper choice of split ratio • As the packets are coming through two different radio links • RLC layer re-ordering logic • Percentage of triple DUPACKS is decreased • Larger congestion window which leads to higher TCP throughputs

43. Performance Results

44. CDF for Different LWIR Bearer Selection Schemes • CDF for all the proposed bearer selection schemes • Min CQI First and Max CQI First are selecting users for steering based on CQI • Min CQI First is more fair as it is serving the cell edge users through Wi-Fi • The Max RLC First is a load-aware scheme which always selects the user with the highest data in its RLC buffer • All users are eligible to be selected for steering which makes it more fair • The other two schemes consider CQI and load in their user selections and thus perform even better • Max RLC First with feedback mechanism achieves better throughput as compared to scheduling • CQI techniques and better fairness as compared to MAX RLC mechanism

45. CDF Result

46. Conclusions and Future Work • LWA • Out-of-order packets problem • LWIR with VWS • Minimizes the delay generally caused in Wi-Fi network • Maximum 85% throughput improvement over the LWA • Only in the downlink • Heavy interaction between the two MAC layers • Only in the co-located scenarios • Future work • Uplink flows • Non-collocated scenario

47. Outline • Overview • LWIP and Wi-Fi Boost Flow Control • LWIR: LTE-WLAN Integration at RLC Layer with Virtual WLAN Scheduler for Efficient Aggregation • A Fair Listen-Before-Talk Algorithm for Coexistence of LTE-U and WLAN • 在OAI平台的LWA網路上使用UDP隧道實作影音串流的雙介面傳送機制 • Summary • References

48. A Fair Listen-Before-Talk Algorithm • Background • Introduction • Coexistence • Listen-Before-Talk (LBT) • LBT Procedure • Fair LBT (F-LBT) Algorithm

49. Background • Recent report predicts that mobile data traffic will hit an annual run rate of 291.6 Exabytes by 2019 • A compound annual growth rate in the data traffic from 2014 to 2019 is 57 percent • Many researchers pay attention to the interworking of different technologies • Long term evolution (LTE) and wireless local area network (WLAN) to • Maximize users’ quality of experience

50. Introduction • LTE-U operation allows seamless data offloading by using a carrier aggregation technique • Since LTE and WLAN are designed to operate in different bands, they do not have any coexistence mechanisms, which leads to significant performance degradation • Since LTE does not sense for channel vacancy prior to transmissions • The interference due to LTE transmissions severely affects the WLAN performance