1/24 VoIP Prioritisation over WLAN for Fixed Mobile Convergence Brendan Kearns, eircom Séan Murphy Performance Engineering Laboratory, University College Dublin
2/24 WLAN Congestion Dominates Voice Quality Light Load Network Core Access Point Network Access Congestion Upstream DELAY Down DELAY Devices Access Point
3/24 Do we really need QOS on WLAN?
4/24 a wasted effort?
5/24 WLAN speed increasing so congestion unlikely? 11Mbit/s b 6 calls 54Mbit/s g 18 calls 100Mbit/s n 40 calls
6/24 In reality, capacity is not so high ….and 5Mbits Max b <<6 calls 28Mbit/s Max g <<18 calls 50Mbit/s Max n <<40 calls
7/24 Use Cases: Home and Enterprise Devices clustered around AP Good channel conditions High utilisation Access Point LaptopRadio Multiple devices give longer wait times Dispersal means lower bitrates Encryption reduces efficiency Congestion at lower utilisation Access Point LaptopRadio The Home The Enterprise
8/24 …...QOS can protect voice quality during traffic congestion.
9/24 … giving reliable voice quality
10/24 Contents Contrast the WLAN access protocol, without and with QOS Test bed description How are voice / data are discriminated Experimental Results Conclusions
11/24 Behaviour without QOS AP & devices contend for access 10 tx time tx busy As we start the AP is transmitting 2 4 Handset, laptop pause for ‘wait time’ Then start random backoff counter tx freeze Laptop gets medium Handset freezes counter Voice packet is delayed AP Upstream DELAY Down DELAY Devices Access Point
12/24 Behaviour without QOS…problems Universal wait and backoff times, thus voice can get trapped behind data traffic all traffic equally degraded in congestion No priority for access point, thus downstream traffic is penalised
13/ operation with QOS e & WMM WMM certification Shorter wait & backoff counter for voice AP brodcasts wait & backoff counter to wireless devices So downstream traffic is prioritised Now handset gets the radio tx wait tx freeze 10 tx time AP Upstream DELAY Down DELAY Devices Access Point
14/24 Test Bed VoIP traffic generation code Tx and Rx are time synchronised Access Point VoIP Call Generator Server Wireless Device Client 1 Call 1 Time Synch Wireless Device Client 8 Call 8 Eight calls sufficient to overload.11b Thus one way delay measured Upstream DELAY Down DELAY Devices Access Point
15/24 QOS Switched off: all traffic is legacy Access Point with WMM Switched Off All Device Types Legacy
16/24 QOS on: four traffic classes Legacy Up Best Effort* Down Voice Down Voice Up 1. Voice Downstream 2. Voice Upstream 3. Best Effort Downstream 4. Best Effort or Legacy Upstream Access Point with WMM Switched On Legacy Device WMM Device
17/24 [EXPT 1] Enterprise: WMM switched off at AP u Calls started at 10 sec intervals u Each call lasts 90 sec u Graph of utilisation Vs time u It shows utilisation building up until seven calls are connected Seventh Call First Call Congestion
18/24 [Expt 1] Enterprise: WMM switched off at AP u Now view from latency perspective u Latency on vertical axis Vs Time u Good quality at six calls u On seventh call u Downstream garbled u Constant 400mS delay u Large packet loss u Upstream minimally impacted Degraded Voice Upstream Latency For a Call Downstream Latency For a call Seventh Call WLAN user cannot hear B party Time sec
19/24 [Expt 2] Enterprise: Some handsets have QOS u Mixture of legacy and WMM calls u Good quality at six calls u On seventh, legacy calls impacted u Upstream garbled u Massive delay u No packet loss u High latency because its got lowest priority u Downstream is not impacted Legacy Downstream Latency Legacy Upstream Latency B party cannot hear WLAN user 7th Call added u Prioritised voice quality is protected
20/24 [Expt 3] Enterprise: File Download u Latency shown for legacy call u Large UDP download for 20 sec u Downstream voice quality degraded u 400ms delay u 20% packet loss u Upstream is not impacted Download Starts WLAN user cannot hear B party Legacy Downstream Latency Legacy Upstream Latency u Repeat expt with prioritised call in progress u Now there is no impact on voice quality
21/24 Summary Downstream LatencyUpstream Latency EXPT 1 No WMM Degraded Voice Constant 400mS Delay Large Packet Loss EXPT 2 Latency for Legacy Devices Competing with Prioritised Voice Degraded Voice Massive Delay No Packet Loss EXPT 3 Latency for Legacy Devices During Data Download Degraded Voice Constant 400mS Delay Large Packet Loss
22/24 Conclusions WMM protects voice quality during congestion Absolute guarantees not possible with WMM QOS case is not compelling for home QOS can deliver significant benifits to the enterprise This is the first published experimental study of QOS on commercially available hardware. It was configured with realistic QOS parameters ……. so the results presented are applicable to real implementations.……... Dziekuje.……...
23/24 Personal details: (1/2) Name: Surname: Company: Phone: Brendan Kearns eircom
24/24 Personal details: (2/2) Name: Surname: Company: Phone: Sean Murphy UCD, Dublin