GPRS優(yōu)化方法、案例培訓(xùn)資料(MOTO)(共113張)

1、Motorola Confidential Proprietary!MOTOROLA and the Stylized M Logo are registered in the US Patent & Trademark Office. All other product or service names are the property of their respective owners. Motorola, Inc. 2002. 中國移動中國移動GPRS無線優(yōu)化課題交流會無線優(yōu)化課題交流會Solution CenterGTSS ChinaMotorola2022年2月14日星期一Moto

2、rola Confidential Proprietary!Page 2q1. GPRS業(yè)務(wù)模型分析業(yè)務(wù)模型分析2022年2月14日星期一Motorola Confidential Proprietary!Page 3USTM considers service activity from the GPRS/UMTS users point of view:(USTM stands for Unified Simple Traffic Models)User ApplicationTelecommunications NetworkUser ApplicationApplication Lev

3、el Service ActivityAt this level the application service behaviour is relatively network implementation independent.However, as the traffic is transported over the network, traffic behaviour within the network will depend on the network implementation and configuration2022年2月14日星期一Motorola Confident

4、ial Proprietary!Page 4As the application interacts with different protocols in the network implementation, application traffic becomes packets.ApplicationApplicationUMTS NetworkUTRANCore NetworkInternetTCP/IPPacket ActivityApplication-level ActivityPacket behaviour may be different in different prot

5、ocol layers, and parts of the network, i.e: UTRAN UMTS Core Network External Packet Data Network TCP & IP packets RLC SDU Radio frames2022年2月14日星期一Motorola Confidential Proprietary!Page 5The traffic characteristics of a service can be very complex. It is often described by a layered data application

6、 model in USTM.The following is a view of the FTP service:SessionBurst(File)Packetttt2022年2月14日星期一Motorola Confidential Proprietary!Page 6v A service session is a logical entity.v Independent of transmission direction.v It is basically described as the period where a user is actively using the servi

7、ce.v During an active service session, the network must maintain connectivity between the users in that session.Sessions are described by:Session Duration (D)Session Inter-arrival time (I)DIt2022年2月14日星期一Motorola Confidential Proprietary!Page 7Bursts described by:Burst Size/Duration ()Inter-burst ti

8、me (T)Ttv The Burst represents a block of real traffic that the application sends/receives.v The Burst is the activity that the bearer channels in the network see.v For many services, a burst equates to the transfer of a file or message.v The burst is uni-directional (uplink and downlink bursts are

9、different).Note: Burst duration may be driven by data size/link speed/load2022年2月14日星期一Motorola Confidential Proprietary!Page 8Telephony:Speech burstsUplinkDownlinkFile burstsUplinkDownlinkPage DownloadsSMS MO:Message BurstUplinkSpeech samples encoded into speech frames.Whilst originator is talking,

10、 speech frame activity generates the speech burst. Unidirectional transmission of the SMS packet. 2022年2月14日星期一Motorola Confidential Proprietary!Page 9SASessionObjectstDItBurststULDLBurstPacketTD - Session durationI - Session inter-arrival time - Burst durationT - Inter-burst timeS - Packet sizeA -

11、Packet inter-arrival time2022年2月14日星期一Motorola Confidential Proprietary!Page 10A PDP context is a logical pipe that provides the transfer of user data across the UMTS network.UMTSExternalNetworkApplicationPDP ContextApplicationFrom either end, the PDP context will look like an IP termination (IP add

12、ress/port) that offers a given level of IP QoS (e.g. acts as a DiffServ code point).Note: The PDP context does not really see the user application, but only sees the IP QoS requirements.UMTSUserLANPDPExt.IPTETETETEUMTS looks like an IP routerReal Network=MS ModemGGSNUEUERemote Host2022年2月14日星期一Motor

13、ola Confidential Proprietary!Page 11v A service session needs a connection across the IP networks to operate.v Thus a service session needs a PDP context over UMTS for its duration.v However, a PDP context may exist before and after a particular service session.v More than one service session can us

14、e a single PDP context.v Different services sessions may share a single PDP context.v PDP contexts may change states (active, standby). This is transparent to the service.Service A sessionsService B sessionsPDP Contextst2022年2月14日星期一Motorola Confidential Proprietary!Page 12Data Collection pointsRANC

15、orePDNGbMDTTPCtcpdump/windumpTcptrace/EtherrealNetsenseGiGb ConsultantRadComGb Link AnalysisEnd-to-End AnalysisPerformance Benchmarkingtcpdump/windumpTcptrace/EtherrealNetsenseEnd-to-End AnalysisEnd User (Data and RF Stats)K1205Gb SnoopGDSScriptsBOTGi Link AnalysistcpdumpWindumpTcptraceEtherrealNets

16、enseTCP AnalyzerTEMS2022年2月14日星期一Motorola Confidential Proprietary!Page 13用戶數(shù)據(jù)流與協(xié)議棧用戶數(shù)據(jù)流與協(xié)議棧用戶數(shù)據(jù)用戶數(shù)據(jù)(IP包包)從終端出發(fā)跨過從終端出發(fā)跨過GPRS承載到達(dá)因特網(wǎng)承載到達(dá)因特網(wǎng), 途中進(jìn)行了大量的協(xié)議封裝和解封裝途中進(jìn)行了大量的協(xié)議封裝和解封裝, 傳輸完成后傳輸完成后, 在在GGSN還原成原來的用戶數(shù)據(jù)還原成原來的用戶數(shù)據(jù)(IP包包), 從從GGSN經(jīng)經(jīng)Gi接口進(jìn)入因特網(wǎng)接口進(jìn)入因特網(wǎng), 我們選擇我們選擇Gi接口作為用戶數(shù)據(jù)流的偵接口作為用戶數(shù)據(jù)流的偵聽和分析點(diǎn)聽和分析點(diǎn).BSS/PCUSGSN

17、因特網(wǎng)因特網(wǎng)GGSNPPPIPTCP/UDPHTTP etc.L2L1IPTCP/UDPHTTP etc.L2L12022年2月14日星期一Motorola Confidential Proprietary!Page 14Gi接口用戶數(shù)據(jù)的封裝接口用戶數(shù)據(jù)的封裝因特網(wǎng)因特網(wǎng)GGSNWap GWGRETCP/UDPOther AppsMail serviceIPTCP / UDP / GREHTTP / WAP / POP3&SMTP / etc.L2L1NetworkTransportApplicationData LinkPhysical2022年2月14日星期一Motorola Confi

18、dential Proprietary!Page 15協(xié)議棧與分析內(nèi)容協(xié)議棧與分析內(nèi)容ApplicationPresentationSessionTransportNetworkData LinkPhysicalApplicationPresentationSessionTransportNetworkData LinkPhysical封裝后的應(yīng)用層報文和實(shí)際傳輸?shù)臄?shù)據(jù), 如FTP DataApplicationApplication應(yīng)用層報文(控制消息等), 如FTP Control在應(yīng)用層只能看到“應(yīng)用消息”, 它被封裝到傳輸層和網(wǎng)絡(luò)層與數(shù)據(jù)一起被傳輸, 在應(yīng)用層的話務(wù)量分析沒有意義.應(yīng)用

19、層做會話分析, 傳輸層做連接分析和話務(wù)分析.傳輸協(xié)議分析應(yīng)用協(xié)議分析應(yīng)用會話分析傳輸連接分析傳輸流量分析網(wǎng)絡(luò)流量分析2022年2月14日星期一Motorola Confidential Proprietary!Page 16分析內(nèi)容ItemsStatistics ValueTrendlineDistributionTraffic Analysis (話務(wù)量分析話務(wù)量分析)網(wǎng)絡(luò)流量分析 GPRS Traffic Bytes Total24 Hour TrendlineDoneNon 杛exmitxActive Subscribers (IP Add#)Active IP address#24 H

20、our TrendlineDoneNetwork Layer Packet size IP Packet sizesniffer givesDoneUL/DL RatioPacket# RatioDoneBytes RatioNetwork Layer ThroughputThroughput in kbps24 Hour TrendlineDone傳輸流量分析 Transport Layer (TCP/UDP/GRE)Sub-totalPie chartDone傳輸協(xié)議分析 Transport Layer Segment sizeTCP/UDP/GREP/ICMP Segment sizeD

21、one應(yīng)用協(xié)議分析 Application Layer TrafficSub-total BytesPie chartDoneSub-total PacketsPie chartRTT Min/max/avg/stddevXSession Analysis (協(xié)議層會話分析協(xié)議層會話分析)傳輸連接分析 Transport Layer (TCP/UDP/GRE)Duration24 Hour TrendlinePie & Bar ChartDoneConversation Number24 Hour TrendlineDoneTraffic (Kbytes)canDoneSession Laye

22、rDuration24 Hour TrendlinePie & Bar Chart (WINS/MSRPC_CL/etc.)Conversation NumbercanTraffic (Kbytes)can應(yīng)用會話分析 Application Layer (HTTP/etc.)Duration24 Hour TrendlinePie & Bar ChartDoneConversation NumbercanDoneTraffic (Kbytes)canDoneHTTP/WAP (應(yīng)用分析應(yīng)用分析)應(yīng)用協(xié)議分析 Throughputkbps24 Hour TrendlineDoneTraffic

23、Mbytes24 Hour TrendlineDoneSessionrefer to upperDoneUser Behavior (用戶行為用戶行為)Avg. traffic per user(CMNET,CMWAP)kbps per IP Add.24 Hour TrendlineDoneApplication vs. Users#User# per ApplicationcanIP Address# Pie Chart DoneUser Online Time per Day?User Traffic per Day?典型用戶定義HTTP UserWAP/MMS UserMail Ser

24、vice UserOther App UserTraffic per Day (Mbytes):2.280.260.280.8Busy Hour:21:0022:0013:00 / 21:0021:0022:0021:0022:00Throughput BH/Max:3.23K/7.19K0.62K0.6K / 1.0K0.82K / 2.15KUsers Number in BH22853084191Users Number in Day142645177501295-PDP Attempt Times:-Online Time per Day:-分析內(nèi)容分析內(nèi)容2022年2月14日星期一M

25、otorola Confidential Proprietary!Page 17qCase Study in 2003 from a European Network2022年2月14日星期一Motorola Confidential Proprietary!Page 18Detailed Analysis GeneralTotal FiguresTotal FiguresAll Sessions+10kByte SessionsTotal Traffic:440MByte330MByteDL/UL Ratio:4,7532,55Sessions:647786936TCP Sessions:8

26、0%100%Session Duration (Avg):10s (85%)1min 50secTraffic per TCP Session (Avg):7,1kB47kBTraffic per UDP Session (Avg):1,1kBN/AAverage Segment Size (DL/UL):320 / N/A Bytes780 / 330 Bytes+10kByte SessionsAll sessions with more than 10kByte traffic in downlink are counted.Source: TCPDump/TCPTrace2022年2月

27、14日星期一Motorola Confidential Proprietary!Page 19Detailed Analysis GeneralApplication Usage (in term of sessions)Source: TCPDump/TCPTraceAll Sessions1% 650 Sessions+10kB Sessions (in DL)1% 70 Sessions2022年2月14日星期一Motorola Confidential Proprietary!Page 20Detailed Analysis GeneralApplication Usage (in t

28、erm of bytes)Source: TCPDump/TCPTraceAll Sessions1% 650 Sessions+10kB Sessions (in DL)1% 70 Sessions2022年2月14日星期一Motorola Confidential Proprietary!Page 21Detailed Analysis GeneralTransport Protocol UsageMajor TrafficTCP HTTP EmailSource: TCPDump/TCPTraceUDPWAPDNSBytesSessionsAll Sessions2022年2月14日星期

29、一Motorola Confidential Proprietary!Page 22Detailed Analysis GeneralPacket Size (UL and DL)Source: TCPDump/NetSenseAll Sessions2022年2月14日星期一Motorola Confidential Proprietary!Page 23Detailed Analysis TCP Parameter Segment Size Requested Applied Window Size Initial, Avg/StdDev/Min/Max Applied Selective

30、 ACK Usage Out of Order PacketsSource: TCPDump/TCPTrace2022年2月14日星期一Motorola Confidential Proprietary!Page 24Detailed Analysis TCP Parameter Selective Acknowledge / Out of Order (on Gi)Source: TCPDump/TCPTrace Selective AcknowledgeAll Sessions+10kByte Sessions (DL) DL:24% 0,04%Used in X% of the TCP

31、Sessions UL:24%2%Used in X% of the TCP Sessions Out of Order Packets DL:0,16%2,93% TCP Sessions with OoO Packets UL:0,07%0,22% TCP Sessions with OoO Packets2022年2月14日星期一Motorola Confidential Proprietary!Page 25Conclusion & RecommendationsqSystem OptimisationqHW/SW DimensioningqFeature Priorization20

32、22年2月14日星期一Motorola Confidential Proprietary!Page 26Conclusion & RecommendationsSystem OptimisationqShort Sessions - General HTTP1.1 is used Causes Low Throughput Low SACK usage for large transfers 3% out of order packet for large transfers on Gi!Recommendation :TCP Stack Tuning ProxySelective ACK,

33、RetransmissionsTCP Client Window Size Tuning (via Driver Installation?)Important for CS3-4Further Analysis on Why so many short (HTTP) sessions?“2022年2月14日星期一Motorola Confidential Proprietary!Page 27Conclusion & Recommendations Hardware/Software DimensioningqSegment Size +60% of packets are +500Byte

34、sRecommendation:Maximum LLC length optimisation (Max. Improvement 1%)qHardware Dimensioning Traffic Distribution, Busy Hour 2pm with 1/3 above minimum2022年2月14日星期一Motorola Confidential Proprietary!Page 28Conclusion & Recommendations Feature PriorizationqSession Duration Sessions are mainly very shor

35、tRecommendation :Overlapping TBF and Rapid Interleaving are very importantqSession Interval Typically below 5seconds. Shows benefit of Super-CoattailRecommendation:SCT Parameters are set well2022年2月14日星期一Motorola Confidential Proprietary!Page 29Conclusion & Recommendations Feature PriorizationqAppli

36、cation Usage 97% of the traffic is TCP UDP traffic is WAP (is connection oriented) and DNSRecommendation : LLC Acknowledgement not necessary2022年2月14日星期一Motorola Confidential Proprietary!Page 30q案例：案例： 引入新應(yīng)用引入新應(yīng)用(Push to Talk)2022年2月14日星期一Motorola Confidential Proprietary!Page 31Case Study Five - In

37、troduction of new application & understanding its impactqCustom Telemetry application was modelled and simulated by GSMSq8 GPRS Timeslots were considered per cellq250, 500 & 750 users in a cell were simulated Packet End-to-End Delay versus Timeslot Allocation: 750 Telemetry Users05010015020025030035

38、0012345678Timeslots (CS1 + CS2 only)Pkt E2E Delay (seconds)Max. E2E delay /secAvg. E2E delay /sec Packet End-to-End Delay versus Timeslot Allocation: 500 Telemetry Users050100150200250300350012345678Timeslots (CS1 + CS2 only)Pkt E2E Delay (seconds)Max. E2E delay /secAvg. E2E delay /sec Packet End-to

39、-End Delay versus Timeslot Allocation: 250 Telemetry Users050100150200250300350012345678Timeslots (CS1+CS2 only)Pkt E2E Delay (seconds)Max. E2E delay /secAvg. E2E delay /sec250 Users500 Users750 Users2022年2月14日星期一Motorola Confidential Proprietary!Page 32New Challenges with PTT over GPRSqPTT is the f

40、irst always-on application to be deployed over GPRS Network PDP context is always on and TBF is setup during conversations could be a capacity constraintqRequirements on GPRS Network Minimal latency Optimal bandwidth to support PTT traffic Cell Reselection & Routing Area Update delays should be mini

41、mal Uplink/downlink TBF release duration should be optimal, to enable PTT conversation to use the same TBF and avoid setting up a new TBF, for every push2022年2月14日星期一Motorola Confidential Proprietary!Page 33Optimise SGSN Ready Timer ValueqIn STANDBY no data reception is possible. To get back to READ

42、Y state, the mobile needs to be paged, thus increasing the set-up time, for the first push. Benefit: By increasing the Ready Timer on the SGSN, the mobiles stay longer in READY state. Drawback:Increased ready timer increases the number GMM signaling messages and causes higher battery consumption.STA

43、NDBYIDLEREADYGPRS attachGPRSdetachREADY timer expiry or forced to STANDBYPDUtrans-mission2022年2月14日星期一Motorola Confidential Proprietary!Page 34Optimised use of PBCCH/PCCH feature MS BTS BSC PCU RF GSL TRAU GDS TRAU GDS RSL GPRS Traffic on CCCH GPRS Traffic on PCCCH PBCCH/PCCCH provides more signalli

44、ng capabilities for both GSM voice and GPRS service by moving all GPRS signalling on to the PBCCH/PCCCH. It reduces internal signalling traffic over the GSL and RSL since there are fewer requests, assignments or paging messages over GSL and RSL. This is due to GPRS signalling traffic shall be carrie

45、d over the TRAU GDS 2022年2月14日星期一Motorola Confidential Proprietary!Page 35Optimised use of PBCCH/PCCH featureBenefits for PTTqIncreased signalling capacity to support PTT type applicationqSeparated signalling for GPRS (PTT) and GSMqFaster pagingqShorter TBF set-up times in heavily loaded networks (c

46、ells) compared to the situation without the PBCCH/CCCH feature.DrawbacksqBy reserving a timeslot for a PBCCH/PCCCH, capacity will be lost for GPRS, though it should be noted that a PBCCH/PCCCH timeslot may still be used for user data (only PPCH & PAGCH i.e. downlink, no uplink). If a cell is RAC bou

47、ndary, the resource on the cell will be used for location area updates and so leave less resource for user data.qNot all existing GPRS MSs support PBCCH and these MSs wont be able to attach when PBCCH is present in the cell2022年2月14日星期一Motorola Confidential Proprietary!Page 36Optimising DRX timerqTh

48、e gprs_drx_timer_max parameter controls how long the mobile station stays in non-DRX mode, when it has left the packet transfer mode and enters the packet idle mode. The maximum value of this parameter is 3 seconds.qBenefit for PTT Decreasing set-up times for consecutive “pushes”, when done within 3

49、 seconds (or whatever duration is set) after last data transmission/reception.qDrawback Optimizing the non-DRX timer is important to increase MS battery lifetime2022年2月14日星期一Motorola Confidential Proprietary!Page 37One Phase accessqOne Phase access is recommended for GPRS Network to reduce TBF set-u

50、p times, this works well for lightly loaded network.qFor a congested network, this may result into RSL congestion. To avoid using the RSL to notify the PCU of an access, the BTS forces the mobile to perform a two-phase access, with the mobile sending a PRR on the reserved block. Since this reserved

51、block is on a PDCH, this signalling is carried out on GPRS TRAU and not the RSL, thereby reducing RSL load. (This requires that uplink blocks are reserved for only PACKET RESOURCE REQUESTS) By reserving blocks for PRRs and forcing two phase packet access RSL congestion is reduced, but the trade off

52、is that two phase packet access results in a much slower TBF establishment time than both one phase and enhanced one phase. 2022年2月14日星期一Motorola Confidential Proprietary!Page 38PTT Performance EnhancementPrepare(GNOS Healthcheck,Analysis &Optimisation,GSMS,Dimensioning)Migrate(Implement GAP Analysi

53、srecommendations,IC&I PTT,Test & Optimise)Monitor(Benchmark PTTperformance, GNOSHealthcheck, Analysis &Optimisation, GSMS)Deliverableis GAPANALYSISDeliverableis PTTImplementedand testedDeliverableis PTTperformanceis alwaysmaintainedat high-performancelevels2022年2月14日星期一Motorola Confidential Propriet

54、ary!Page 39qGPRS網(wǎng)絡(luò)無線優(yōu)化原則及案例分析網(wǎng)絡(luò)無線優(yōu)化原則及案例分析2022年2月14日星期一Motorola Confidential Proprietary!Page 40GPRS 網(wǎng)絡(luò)結(jié)構(gòu)2022年2月14日星期一Motorola Confidential Proprietary!Page 41SGSNGi頻率計劃UmConf天線PCUPIXDNSCom-HubGGSNGnGbDHCPNTPOMC-ROMC-G NTOMC-G UnixISSGPRS優(yōu)化所涉及的網(wǎng)絡(luò)單元GrFRAME路由協(xié)議隧道協(xié)議FRLLC/MACTCP/IPGs2022年2月14日星期一Motorol

55、a Confidential Proprietary!Page 42AirIPCircuitNetwork CMCC主要統(tǒng)計主要統(tǒng)計 終端應(yīng)終端應(yīng)用測試用測試路測路測數(shù)據(jù)數(shù)據(jù)GPRS優(yōu)化所涉及的改善指標(biāo)2022年2月14日星期一Motorola Confidential Proprietary!Page 43GPRS 的優(yōu)化方式應(yīng)用層應(yīng)用層傳輸層傳輸層MSPCUBTSBSCCorporateVPNWAPGatewayRadiusDHCPDNSFRBillingLIANSGSNGGSNISS2022年2月14日星期一Motorola Confidential Proprietary!Page 4

56、4GPRS 測試環(huán)境Maxis IntranetAntennaInternetGGSN100bT HubRadius- Server(Solaris)Benchmark Server(Solaris)RSM1RSM2CommHubGn-Desktop(Win2000)SGSNHP37907PCUVLAN 20GnLaptop-Test execution(Win2000)Laptop-MDTT(Win2000)VLAN 10GGSN-GiVLAN 30WANBorderRouterGi- Trace:Tool: tcpdumpBmServer-Logs/Trace:Tool: tcpdump

57、log-filesGn- Trace:Tool: windumpSGSN-Logs:Tool: log-filesGb- Trace:Tool:HP37907PCU- Logs:Tool: PCU-LogsMS- Logs/Trace:Tool: log-file CommView Perf. MonitorMDTT-Logs:Tool: MDTTMobileOMC-RCall Trace- Logs:Tool: IOS CTP設(shè)備鏈路設(shè)備鏈路測試鏈路測試鏈路2022年2月14日星期一Motorola Confidential Proprietary!Page 45GPRS 無線優(yōu)化原則數(shù)據(jù)分

58、析E2E性能通過DT/CQT對終端性能分析，包括：qAttach success rate;qPDP active success rate;qRAU success rate;qCS usage;qCell reselect;qNetwork latency, throughput, bandwith;qBLER2022年2月14日星期一Motorola Confidential Proprietary!Page 46GPRS 無線優(yōu)化原則數(shù)據(jù)分析統(tǒng)計方向q系統(tǒng)容量分析PCU/PRP load分析；熱點(diǎn)地區(qū)GPRS時隙分配；Switch時隙對GSM影響；q上下行吞吐量分析；qBLER分析；q

59、擁塞分析；q各信令層丟包分析；2022年2月14日星期一Motorola Confidential Proprietary!Page 47GPRS 無線優(yōu)化原則數(shù)據(jù)分析硬件告警分析qPCU，RXCDR，BSC，BTS，TCU告警分析；q傳輸信令告警；q天饋線系統(tǒng)； 2022年2月14日星期一Motorola Confidential Proprietary!Page 48GPRS 無線優(yōu)化原則數(shù)據(jù)分析數(shù)據(jù)庫分析q功率控制分析；q小區(qū)重選分析；q鄰區(qū)拓?fù)浣Y(jié)構(gòu)分析；qGPRS接入及信令Timer;2022年2月14日星期一Motorola Confidential Proprietary!Page

60、 49GPRS 無線優(yōu)化原則數(shù)據(jù)分析無線環(huán)境分析q上下行質(zhì)量分布；q上下行電平分布；qC/I分布；qPathbalance分布；qTA分布；2022年2月14日星期一Motorola Confidential Proprietary!Page 50GPRS 無線優(yōu)化原則問題總結(jié)系統(tǒng)問題總結(jié)q系統(tǒng)硬件告警及問題q頻率干擾問題q覆蓋問題q數(shù)據(jù)庫問題q信令配合q系統(tǒng)資源問題2022年2月14日星期一Motorola Confidential Proprietary!Page 51GPRS 無線優(yōu)化原則問題處理無線環(huán)境優(yōu)化q系統(tǒng)C/I、BLER分析q頻率調(diào)整q覆蓋/天線調(diào)整q相鄰小區(qū)優(yōu)化2022年2月