Slide title In CAPITALS 50 pt Slide subtitle 32 pt Active Queue Management for LTE uplink in eNodeB Yifeng Tan Supervisor: Professor Riku Jäntti Instructor:

Similar presentations


Presentation on theme: "Slide title In CAPITALS 50 pt Slide subtitle 32 pt Active Queue Management for LTE uplink in eNodeB Yifeng Tan Supervisor: Professor Riku Jäntti Instructor:"— Presentation transcript:

1 Slide title In CAPITALS 50 pt Slide subtitle 32 pt Active Queue Management for LTE uplink in eNodeB Yifeng Tan Supervisor: Professor Riku Jäntti Instructor: D.Sc. Riikka Susitaival

2 Top right corner for field-mark, customer or partner logotypes. See Best practice for example. Slide title 40 pt Slide subtitle 24 pt Text 24 pt Bullets level 2-5 20 pt Ericsson Confidential2009-02-202 Long-Term Evolution (LTE)  Evolved 3G Radio Access Network  Provides high data rate –Downlink: up to 300 Mbps –Uplink: 50 Mbps  ”Mobile Internet” –TCP/IP over wireless R99Rel4Rel5Rel6 4 Rel7Rel8 WCDMAHSDPAHSPAHSPA evolution LTE

3 Top right corner for field-mark, customer or partner logotypes. See Best practice for example. Slide title 40 pt Slide subtitle 24 pt Text 24 pt Bullets level 2-5 20 pt Ericsson Confidential2009-02-203 TCP basics  One of the core protocols of the Internet  Provides reliable transmission –Automatic Repeat reQuest (ARQ): Receiver sends ACKs to indicate the reception of segments.  TCP sender’s window: How many packets are ”in flight” –Ideally: Sender’s window = Pipe Capacity (Bandwidth ∙ RTT) –Smaller: under-utilization of the link –Bigger: queuing of packets. When queue exceeds buffer limit, some packets have to be discard → congestion

4 Top right corner for field-mark, customer or partner logotypes. See Best practice for example. Slide title 40 pt Slide subtitle 24 pt Text 24 pt Bullets level 2-5 20 pt Ericsson Confidential2009-02-204 TCP Congestion Control  TCP sender tries to ”probe” optimal send-window –ACK received →Increase send-window ”More bandwidth available” –Packet loss → Decrease send-window ”Light congestion” –Time-outs → Start probing from scratch ”Serious congestion”

5 Top right corner for field-mark, customer or partner logotypes. See Best practice for example. Slide title 40 pt Slide subtitle 24 pt Text 24 pt Bullets level 2-5 20 pt Ericsson Confidential2009-02-205 TCP Congestion Control example

6 Top right corner for field-mark, customer or partner logotypes. See Best practice for example. Slide title 40 pt Slide subtitle 24 pt Text 24 pt Bullets level 2-5 20 pt Ericsson Confidential2009-02-206 Queue Management  Passive Queue Management –Drop packets when the buffer is full –We can only decide ”which” but not ”when” –Drop-tail, drop-front, random-drop –PQM may cause some problems  Significant end-to-end delay  Unfair sharing  Viscous web browsing  Active Queue Management –Drop packets even there is space –We can decide both ”which” and ”when”

7 Top right corner for field-mark, customer or partner logotypes. See Best practice for example. Slide title 40 pt Slide subtitle 24 pt Text 24 pt Bullets level 2-5 20 pt Ericsson Confidential2009-02-207 Advantages of AQM  Maintains lower average queue size as well as end-to- end delay than PQM –Important when e.g. Web browsing  Prevents lock-out –Avoid the situation where one or a few TCP flows dominate the link  Reduce the number of packet drops –There is enough space to accommodate the packets in burst

8 Top right corner for field-mark, customer or partner logotypes. See Best practice for example. Slide title 40 pt Slide subtitle 24 pt Text 24 pt Bullets level 2-5 20 pt Ericsson Confidential2009-02-208 Data transmission in LTE uplink  Buffer Status Report (BSR): created by UE and sent to eNodeB to indicate the queue size in UE’s buffer UE BSR Data

9 Top right corner for field-mark, customer or partner logotypes. See Best practice for example. Slide title 40 pt Slide subtitle 24 pt Text 24 pt Bullets level 2-5 20 pt Ericsson Confidential2009-02-209 Example: PDCP discard  Specified in 3GPP TS 36.323 V8.2.1 for each outgoing packet (PDCP SDU) if (delay > delayThreshold ) discard packet else transmit packet  Suffer throughput degradation UE BSR Data PDCP discard

10 Top right corner for field-mark, customer or partner logotypes. See Best practice for example. Slide title 40 pt Slide subtitle 24 pt Text 24 pt Bullets level 2-5 20 pt Ericsson Confidential2009-02-2010 Transmitter-AQM Algorithm for each outgoing packet if(size <= lowerDropThreshold) transmit packet else if (delay > minAgeThreshold AND (now-previousDropTime > minInterDropTime)) OR (delay >maxAgeThreshold)) discard packet previousDropTime=now else transmit packet UE channel RLC eNodeB Drop triggered? No Yes PDCP PDUs RLC SDU RLC PDU  Improve the throughput of PDCP discard  Maintains congestion window bigger than PC –> minAgeThreshold  Prevent from consecutive packet drops -> minInterDropTime  Prevent from drop when queue size is small -> lowerDropThreshold  Drain buffer fast -> maxAgeThreshold

11 Top right corner for field-mark, customer or partner logotypes. See Best practice for example. Slide title 40 pt Slide subtitle 24 pt Text 24 pt Bullets level 2-5 20 pt Ericsson Confidential2009-02-2011 Implement in eNodeB?  Why? –Transmitter-AQM was suggested to be standardized but not accepted –UE with or without AQM may have quite different performance –We may control the UE buffer from eNodeB by Receiver-AQM so that all the UEs can have improved throughputs  We need to know – queue size (BSR) – packet delay -> How? UE BSR Data R-AQM

12 Top right corner for field-mark, customer or partner logotypes. See Best practice for example. Slide title 40 pt Slide subtitle 24 pt Text 24 pt Bullets level 2-5 20 pt Ericsson Confidential2009-02-2012 A method to estimate queuing delay  The method estimate the delay of headmost data in a queue  some assumptions 1.We can monitor the queue length (Q i ) discontinuously with small time intervals 2.The amount of data left the queue in each time interval is known (L i ) 3.The length of time interval is known (∆t i )  The amount of new incoming data (R i ) in each time interval can be calculated by R i = Q i + L i – Q i-1

13 Top right corner for field-mark, customer or partner logotypes. See Best practice for example. Slide title 40 pt Slide subtitle 24 pt Text 24 pt Bullets level 2-5 20 pt Ericsson Confidential2009-02-2013 Waiting time Queue size ∆t 1 /2 0Q1 Time t0 0 Time t1 Time t2 Q1 + R2R2 ∆t 1 /4+∆t 2 ∆t2∆t2 Q2 ∆t 1 /2+∆t 2 At the beginning...

14 Top right corner for field-mark, customer or partner logotypes. See Best practice for example. Slide title 40 pt Slide subtitle 24 pt Text 24 pt Bullets level 2-5 20 pt Ericsson Confidential2009-02-2014 Waiting time Queue size 0 0 RnRn ∆tn∆tn ∙∙∙∙∙∙ At Time tn Q n-1 + R s QnQn

15 Top right corner for field-mark, customer or partner logotypes. See Best practice for example. Slide title 40 pt Slide subtitle 24 pt Text 24 pt Bullets level 2-5 20 pt Ericsson Confidential2009-02-2015 How to implement in eNodeB  Queue length (Q i ) -> BSR  ∆t i -> time interval of receiving BSRs  Leaving data (L i ) -> the data delivered to higher layer in eNodeB during each ∆t i

16 Top right corner for field-mark, customer or partner logotypes. See Best practice for example. Slide title 40 pt Slide subtitle 24 pt Text 24 pt Bullets level 2-5 20 pt Ericsson Confidential2009-02-2016 Estimated delay vs. Real delay

17 Top right corner for field-mark, customer or partner logotypes. See Best practice for example. Slide title 40 pt Slide subtitle 24 pt Text 24 pt Bullets level 2-5 20 pt Ericsson Confidential2009-02-2017 Receiver-AQM Algorithm for each outgoing packet if(BsrSize <= lowerDropThreshold) transmit packet else if (estimatedDelay > minAgeThreshold AND (now - previousDropTime> minInterDropTime)) discard packet previousDropTime=now else transmit packet Drop triggered? RLC UE eNodeB PDCP PDUs RLC SDU No Yes channel UE BSR Data R-AQM  Drop packets in eNodeB  The algorithm is similar to T-AQM –Queue size reported by BSR -> real queue size –Estimated delay -> real delay

18 Top right corner for field-mark, customer or partner logotypes. See Best practice for example. Slide title 40 pt Slide subtitle 24 pt Text 24 pt Bullets level 2-5 20 pt Ericsson Confidential2009-02-2018 Performance comparison with one TCP flow

19 Top right corner for field-mark, customer or partner logotypes. See Best practice for example. Slide title 40 pt Slide subtitle 24 pt Text 24 pt Bullets level 2-5 20 pt Ericsson Confidential2009-02-2019 Performance comparison with multiple TCP flows

20 Top right corner for field-mark, customer or partner logotypes. See Best practice for example. Slide title 40 pt Slide subtitle 24 pt Text 24 pt Bullets level 2-5 20 pt Ericsson Confidential2009-02-2020 Performance if both AQM implemented (Bandwidth = 3 Mbps)

21 Top right corner for field-mark, customer or partner logotypes. See Best practice for example. Slide title 40 pt Slide subtitle 24 pt Text 24 pt Bullets level 2-5 20 pt Ericsson Confidential2009-02-2021 Some conclusions  Using AQM can efficiently reduce the end-to-end delay in LTE uplink.  PDCP discard may cause considerable throughput reduction in some situations.  R-AQM and T-AQM can maintain small delay as well as high throughput in most situations.  R-AQM is a good option to enhance the performance of UE from network side, no matter if the UE has implemented any AQM itself.

22 Top right corner for field-mark, customer or partner logotypes. See Best practice for example. Slide title 40 pt Slide subtitle 24 pt Text 24 pt Bullets level 2-5 20 pt Ericsson Confidential2009-02-2022


Download ppt "Slide title In CAPITALS 50 pt Slide subtitle 32 pt Active Queue Management for LTE uplink in eNodeB Yifeng Tan Supervisor: Professor Riku Jäntti Instructor:"

Similar presentations


Ads by Google