Network Selection Optimization

1. Network Selection Optimization Authors: Date:2011-07-20 Gabor Bajko, Nokia, Canpolat, Intel, Emmelmann, FOKUS

2. Abstract An optimized network discovery and selection is presented. Today, a STA requiring network information has to conduct AP discovery (scanning) and network discovery. The overhead occurring in environments with a large number of APs or with several STAs simultaneously starting AP/network discovery can be reducing by combing both phases as presented herein. Gabor Bajko, Nokia, Canpolat, Intel, Emmelmann, FOKUS

3. What we have today: one STA several APs SSID3 SSID1 SSID2 PROBE.request (Broadcast) PROBE.response AP discovery (alternative: wait for all beacons) PROBE.response PROBE.response ANQP (Table 8-176 infoIDreq) ANQP (Table 8-176 infoIDreq) ANQP (Table 8-176 infoIDreq) Network discovery ANQP Resp (ANQP IE) ANQP Resp (ANQP IE) ANQP Resp (ANQP IE) • - ANQP is unicast • Each SSID is queried until a suitable network is found. • This frame exchange is conducted for every STA that wants to conduct a link set-up Gabor Bajko, Nokia, Canpolat, Intel, Emmelmann, FOKUS

4. Let • AP – number of APs in communication range • STA – number of STAs “simultaneously” requesting link set-up(simultaneously = within a given, short timeframe) • Then • AP-Discovery requires: STA * (1 + AP)message exchanges • Network Discovery requires in between 1st AP is selected: STA * 2 and last AP is selected: STA * 2 * APmessage exchanges What we have today: cost of AP discovery & network discovery 1 Probe.req + 1 probe.res for each AP Done for each STA Done for each STA 1 ANQP.req+ 1 1 ANQP.res Done for every AP Gabor Bajko, Nokia, Canpolat, Intel, Emmelmann, FOKUS

5. Pro: low processing at AP (parsing / evaluation done at STA)  fast Con: Large overhead, STA #2 couldhaveobtainedtheanswerfromthecontents of the 1st reply How ANQP is done today ANQP (Table 8-176 infoIDreq – ask for property # 1) ANQP Resp (ANQP IE -- all properties including property #1 are returned) ANQP (Table 8-176 infoIDreq – ask for property # 2) Same information containedin the response. ANQP Resp (ANQP IE -- all properties including property #2 are returned) Gabor Bajko, Nokia, Canpolat, Intel, Emmelmann, FOKUS

6. Optimized approach Same AP Discovery as we have today  each STA issues a probe.req. ANQP (Table 8-176 infoIDreq – ask for property #1) broadcast ANQP Resp (ANQP IE – reply with all properties)  broadcast ANQP Resp (ANQP IE – reply with all properties)  broadcast ANQP Resp (ANQP IE – reply with all properties)  broadcast STA 1 gets answer it requested STA 2 parses answers as well  high properbilitythat STA 2 doesnotneed to request additional information Gabor Bajko, Nokia, Canpolat, Intel, Emmelmann, FOKUS

7. Remember what we have today: 1st AP is selected: STA * 2 and last AP is selected: STA * 2 * AP • With improved network discovery: • Best case (all STAs can retrieve required properties from 1st response first AP is selected: 1 + AP • Worst case (no STA can retrieve required properties from 1st response last AP is selected: STA * ( 1 + AP ) • Gains: • Worst case performance of improved scheme has same performance as expected mean of message exchange of today’s scheme • But we can do even better: Since responses are broadcast as well, why not combine network discovery and AP discovery phase ? How much can we save in Network Discovery? Done for each STA 1 ANQP.req+ 1 1 ANQP.res Done for every AP ANQP.req. ANQP.res from every AP Gabor Bajko, Nokia, Canpolat, Intel, Emmelmann, FOKUS

8. Proposed optimized network selection procedure SSID3 SSID1 SSID2 Beacon (interworking capability) ANQP (Table 8-176 infoIDreq) broadcast • - ANQP to be sent to the broadcast address • AP answers if it can satisfy the condition in the request • ANQP responses are sent unicast • Saves (n-1)*RTT during the network selection procedure, n = nr of SSIDs queried, and the passive scanning period ANQP Resp (ANQP IE) ANQP Resp (ANQP IE) Network Selection (SSID2) Association Request Gabor Bajko, Nokia, Canpolat, Intel, Emmelmann, FOKUS

9. Optimized approach II: combine with (active) network discovera X X X X X Same AP Discovery as we have today  each STA issues a probe.req. ANQP (Table 8-176 infoIDreq – ask for property #1) broadcast ANQP Resp (ANQP IE – reply with all properties)  broadcast Include informationfrom probing in “new”message ANQP Resp (ANQP IE – reply with all properties)  broadcast ANQP Resp (ANQP IE – reply with all properties)  broadcast Gabor Bajko, Nokia, Canpolat, Intel, Emmelmann, FOKUS

10. AP-Discovery requires: STA * (1 + AP) • New, improved Network Discovery requires in between • Best case (all STAs can retrieve required properties from 1st response last AP is selected: 1 + AP • Worst case (no STA can retrieve required properties from 1st response last AP is selected: STA * ( 1 + AP ) Further improvement: X X X X X 1 Probe.req + 1 probe.res for each AP Done for each STA ANQP.req. ANQP.res from every AP Gabor Bajko, Nokia, Canpolat, Intel, Emmelmann, FOKUS

11. Obviously not  • E.g., in an enterprise environment, STAs might not need to do any network selection at all •  use „old“ scheme • Employ novel / combined scheme as an alternative scheme to conduct a link set-up • STAs can decide / be told (via MIB variable ?) which scheme to use Should this approach always be used? Gabor Bajko, Nokia, Canpolat, Intel, Emmelmann, FOKUS

12. Is there interest in the group to receive further presentations and draft amending text based on the ideas presented afore? Yes: No: Straw Poll Gabor Bajko, Nokia, Canpolat, Intel, Emmelmann, FOKUS