Smart Mobile Cloud NetSys - KiVS 2013 Bernd Haberland Alcatel-Lucent Bell Labs COPYRIGHT © 2011 ALCATEL-LUCENT. ALL RIGHTS RESERVED. ALCATEL-LUCENT — INTERNAL PROPRIETARY — USE PURSUANT TO COMPANY INSTRUCTION C-RAN Requirements (1) • OPEX and CAPEX reductions • reduction of BBU sites • taking benefit of pooling gains achievable in a cluster of radio sites (range 50-150) • Pooling is done on data Radio Bearer Level (more efficient as on cell level) • Support of LTE advanced as CoMP, HetNet (ICIC) and Carrier Aggregation • Low latency scheduler coordination & data transfer to joint processing points within a BBU pool • Reducing blocking probabilities caused by a lack of BBU processing resources • by offloading to remote processors • Flexibility to allocate BBU processing resources from low traffic areas to high traffic areas • On Intra BBU pool level • On Inter BBU pool level • The vision: to support heterogeneous traffic profiles over time (enterprise, week-end events, stadium etc.) within a radio site cluster or between radio cluster by one or different BBU pools and guarantee a better utilization of the installed processing resources • C-RAN Requirements (2) • Energy saving aspects •Switching off BBU processors in low traffic load situations • Dynamic Spectrum sharing in Multi-standard context •Semi-static Re-arrangement of BBU modems by software replacement between 3G and 4G • Recovery of failures •A BBU processing board failure should not cause the loss of a cell or site • A failure of the Decentralized Cloud Controller should not cause the loss of a radio site cluster Mobile Cloud Overall Architecture HTN Cluster HTN Cluster Metro Metro RRH+ Metro Metro RRH+ RRH RRH … eX2 S1 Iub Abis S1 Iub Abis DCC BBU RRH: Remote Radio Head eX2: Enhanced X2 interface (Ring only as example) MSSeX2 DCC: Decentralized Cloud Controler RRH MSS- S1 Iub Abis DCC BBU RRH … Metro Metro RRH+ eX2 BBU BBU MSS-BBU: Multi-site/standard BBU eX2 DCC DCC HTN: Heterogeneous Network HTN Cluster RRH eX2 RRH eX2 MSS- … RRH RRH RRH MSS- RRH RRH RRH eX2 MSSBBU eX2 DCC S1 Iub Abis In the Cloud (inter MSS-BBU) • on User level • on Cell level S1 Iub Abis Challenge: • Low latency transport MSS-BBU to RRH & inter MSS-BBU Mobile Cloud General Backhaul Architecture RRH eNB RRH eNB RRH eNB RRH eNB RRH eNB RRH eNB RRH eNB RRH eNB RRH eNB RRH eNB RRH eNB MSSLast mile DCC BBUrouter Last mile router RRH eNB MSSLast mile DCC BBUrouter S1 + eX2 S1 + eX2 MSSRouter DCC BBU Router S1 5 COPYRIGHT © 2011 ALCATEL-LUCENT. ALL RIGHTS RESERVED. Last mile router MSS-BBU Functional Architecture cell S1 MME eNB Control Functions RRH1 Scheduler RB11 UP antenna RRH3 RB221 UP UP MUX/ MUX/ MUX/ MUX/ DEMUX DEMUX DEMUX DEMUX S1 - U RB31 UP PHYcell RB41 UP NB Control Functions Iub BTS Control Functions LTE(FDD/ TDD) RRH4 RRH5 antenna Spreader/ Spreader/ Spreader/ Spreader/ Spreader/ Despr Despr Despr Despread Despr RB51 UP Interface II MACMACMAC(e)hs/e (e)hs/e (e)hs/e Scheduler Scheduler Scheduler UP j-1 UP i UP j cell Cell Signals eX2 To other MSSBBUs DCC UMTS UP: User Processing stack (from S1 termination to PHYuser) RRHm PHYcell UP i-1 Abis RRH2 eX2: Enhanced X2 6 COPYRIGHT © 2011 ALCATEL-LUCENT. ALL RIGHTS RESERVED. ALCATEL-LUCENT — INTERNAL PROPRIETARY — USE PURSUANT TO COMPANY INSTRUCTION Traffic scaling of eNodeB functions UP, Scheduler, Cell Control functions Functions Dependencies S1 Termination (User Plane) # data radio bearers S1 data rate PDCP ROHC # data radio bearers S1 data rate Configuration parameters for the air interface (type of HC) PDCP ciphering # data radio bearers S1 data rate Configuration parameters for the air interface (security algo) RLC_MAC # data radio bearers Configuration parameters for the air interface (RLC mode) Radio conditions concatenation, segmentation and reassembly PHYuser # data radio bearers Scheduler # data radio bearers Dedicated control # data radio bearers Cell control Constant PHYcell Constant MAC-PHY rate Radio conditions PHYII rate Radio conditions resource allocation Possible PHY Interfaces From MSS-BBU to RRH MAC-PHY S1Termination/ Termination/ S1 S1 Termination/ PDCP/RLC/MAC PDCP/RLC/MAC PDCP/RLC/MAC I FEC FEC FEC QAM++ QAM QAM + multimultimultiantenna II antenna antenna mapping mapping mapping signal flow (DL) IFFT + CPin Resource III mapping P/S BB to RF to RRH encode to CPRI User processing IV or IV‘ Cell processing MAC/RLC/PDCP MAC/RLC/PDCP MAC/RLC/PDCP S1 Termination S1Termination Termination S1 -1 + -1 QAM -1 QAM + -1 FEC -1 FEC -1 FEC MAC-PHY I QAM + multimultimultiII antenna antenna antenna Processing Processing Processing (e.g.MRC) MRC) (e.g. MRC) (e.g. Resource demapping III CPout + FFT S/P Interface Naming RF to BB decode from CPRI IV or IV‘ from RRH signal flow (UL) PHY (I) Soft-bit fronthauling (softbits + control info) (II) Subframe data fronthauling (frequency domain I/Q + control info) (III) Subframe symbol fronthauling (frequency domain I/Q) (IV) CPRI fronthauling (time domain I/Q) (IV‘) Compressed CPRI fronthauling (time domain I/Q) 8 COPYRIGHT © 2011 ALCATEL-LUCENT. ALL RIGHTS RESERVED. ALCATEL-LUCENT — INTERNAL PROPRIETARY — USE PURSUANT TO COMPANY INSTRUCTION UP Data Rates From MSS-BBU to RRH Site (DL) • Links from MSS-BBU to Macro site (Downlink) • Each Macro site holds 3 sectors • Transmission bandwidth of Macro site as given below • LTE-FDD, 4 antennas, 20 or 40 MHz • Link level protocol not yet included on interfaces I,II and III • Process split at several potential interfaces [Gb/s] per site IV IV' III II/100% II/30% I/100% I/30% MACPHY*/ 100% Data rate (40 MHz BW, Gb/s) 29,5 10,9 5,6 5,4 2,5 0,53 0,19 0,24 sites/ 10 Gb/s link 0 0 1 1 4 18 53 41 Data rate (20 MHz BW, Gb/s) 14,7 5,5 2,8 2,7 1,25 0,27 0,09 0,12 0 1 3 3 8 37 111 83 sites/ 10Gb/s link *average spectral efficiency: 2,0 b/s/Hz 9 COPYRIGHT © 2011 ALCATEL-LUCENT. ALL RIGHTS RESERVED. Pros and Cons of potential PHY fronthauling interfaces Pros Cons MAC-PHY Very low bandwidth needed Dependent on S1 data rate Pooling only for L2/L3 All PHY functions to RRH site Several C-RAN features not supported PHY I Low bandwidth needed Load dependent Pooling only for L2/L3 and FEC All other PHY functions to RRH C-RAN features only partly supported No latency improv. for CoMP In-band protocol needed for Modulation, Multi-antenna proc. and PRB allocation PHY II Moderate bandwidth needed (Load dependent) Pooling for all user related functions from L3-L1 Low latency support of CoMP PHY cell functions to RRH In-band protocol for PRB allocation PHY III Moderate bandwidth needed (Load independent) Same other pros as for PHYII No in-band protocol for PRB allocation needed Remaining PHY cell functions as IFFT, FFT,CPI,CPR to RRH PHY IV’ All pros of PHYIII High bandwidth needed Additional processing effort for CPRI compression at BBU pool and RRH PHY IV All pros of PHYIII Deployed RRHs can be reused Very high bandwidth needed PHY II or PHY III could be a good choice !!! Note: all interfaces need low latency transport interface (Scheduler within MAC) Virtualization Principle of a UP (example LTE) MME/SGW S1-MME S1-U DCC Router Cell Control Functions BBU2 BBU1 MSS-BBU internal connection UP1…UPx Inter MSSBBU via eX2 IF UPx+1… UPn DCC Scheduler PHY cell MSS-BBU1 BBU3 Inter MSS-BBU via eX2 IF UPn+1… UPn+y MSS-BBU2 Some definitions: - For the users 1..x the BBU1 is the Home and the Serving BBU - For the users x+1..n the BBU1 is the Home BBU and the BBU2 the Remote BBU - For the users n+1…n+y the BBU1 is the Home BBU and the BBU3 the Remote BBU 11 COPYRIGHT © 2011 ALCATEL-LUCENT. ALL RIGHTS RESERVED. ALCATEL-LUCENT — INTERNAL PROPRIETARY — USE PURSUANT TO COMPANY INSTRUCTION Processing split for one cell for SOC realization within one BBU Pool and comparison with GPP configurations BBU2 BBU3 UPx+1 …UPm BBU1 UP 1..x Cell functions PHYcell (MUX) Scheduler UPm+1…UPo Ethernet Switch + DCC S1 IF I or II to RRH • It is assumed that BBU-SOCs can include various technologies such like GPPs, DSPs or HW Accelerators • No independent scaling depending on traffic possible for the parts. • MAC + Upper Layer on GPP • PHY on DSP and HW Accelerators • complete UP allocation to the BBUx for virtualization makes sense • GPP based configurations offer more design flexibility: •Independent pools for MAC + Upper Layer and PHYuser with separate virtualization Architecture and Processing split for GPP configurations within one MSS-BBU Pool GPP configuration approach 1,2,3: Different Scheduler Positions GPP++HW HWaccelerator acceleratorpool pool GPP GPP + HW Accelerator pool PHYuser PHYuser PHYuser MUX MUX PCIE-IB GPPpool pool GPP GPP pool MAC+ Upper Layers MAC+ Upper Layers MAC+ Upper Layers Scheduler CellScheduler Control Cell Control Cell PCIE-IBControl Ethernet++IB-Switch PCIe Switch + DCC Ethernet + DCC S1, eX2 PCIE-IB HWAccelerator Acceleratorpool pool HW HWPHYcell Acc. config. (MUX IF II) PHYcell (MUX IF II) PHYcell (MUX IFControl II) PHYCommon Common Channels PHY Control Channels PHY Common Control Channels IF I or II to RRH Fixed configuration scales with number of cells/sites Possible MSS-BBU HW Architecture (with SOC-BBUs) Radio Radio Site k+1 Site k+2 XFP XFP Radio Site m Radio Site 1,2 XFP XFP XFP Radio Site 3,4 Radio Site k -1,k Radio Site 5,6 XFP XFP XFP XFP XFP XFP Fronthauling Switch Cell Signal CFM CFM CFM BBU1 BBU2 BBU3 (from other MSS -BBU) BBUk eX2 IF eX2 UP & Control User Level (from/to other MSS -BBU) Ethernet Switch/ Infiniband DCC Router/Address Dispatcher LTE+ GSM WCDMA + GSM Control Abis Iub S1 -U S1 -MME SRIO • CFM (Cell Function Module): includes a set of cell functions: Cell control part and PHYcell • needed for BBU HW red. (benefit from pooling gains), Recovery (BBU failure), energy saving • Fronthaul Switch: •needed for recovery (BBU failure), energy saving, inserting cell signals from neighbor pools Control Flow Architecture RBC: Radio Bearer Control LRM: Local Resource Manager Resource Pooling Management Algorithms (Intra RM) • Measurements • link bandwidth, link latency • used processing capacity • metric PHYuser: • Processing effort /radio bearer setup request = f (PRB usage, MCS, no. of antennas for TX/RX diversity or MiMO) • Total processing effort= SUM (processing effort /per bearer setup request) • metric for MAC and upper layers: Processing effort = f(S1 rate per radio bearer setup request, no of radio bearer setup requests) • Processing Resource prediction • use Radio-Bearer_Setup_Request information • QoS handling • take into account dynamic Air interface behavior • operator policies handling • Resource Placing decision • process the cost as a function of: • needed processing resource, compared with available processing capacity • link latency and bandwidth • operator policies w.r.t. traffic distributions and energy saving aspects • Connection management and instantiation of user processing functions • with Open Flow Control (look-up table) or per routing header sequence System Simulation study parameters (simulation and simulation results contributed from Dipl.-Ing. Thomas Werthmann from the Institute of Communication Networks and Computer Engineering in Stuttgart, Germany) Parameter Value Scenario LTE, 10 MHz uplink, 10 MHz downlink FDD, 3GPP 25.814 Macro Case 1 with some modifications Playground 4.1 km2, 19 sites, 500 m distance, 3 sectors per site Mobiles Unlimited number, not moving during transmission, channel measurement and reporting, uplink transmit power 23 dBm Channel model Pathloss and static shadowing Handover Based on SINR Outage Requests from mobiles dropped if SINR below -3,9 dB Scheduling Cell reuse one scheme, frequency-diverse and model for frequency-selective and proportional-fair bandwidth assignment [Ellenbeck] Link adaptation Modulation and coding scheme with highest spectral efficiency selected Load distribution Randomly according to a configurable mix of uniform & Gaussian, mean in central site; non-full buffer model Admission Control The 100 % reference load for each user distribution corresponds to the load at which 1% of the requests are dropped by admission control Request messages Created on application layer according to a Poisson inter-arrival process; varying object sizes according to a mix of 3 log-normal distributions, scaled by factor 0.5 [Hernandez-Campos] Response messages One per request; varying object sizes according to a mix of 3 log-normal distributions [HernandezCampos], maximum object size limited to 100 MB Processing resources Processing resources scale according to a processing resource model; for PHYuser part F( #PRBs, MCS, #spatial antennas) [Ellenbeck] Jan Ellenbeck, Johannes Schmidt, Ulrike Korger, Christian Hartmann: A Concept for Efficient System-Level Simulations of OFDMA Systems with Proportional Fair Fast Scheduling, GLOBECOM Workshops 2009 [Hernandez-Campos] F. Hernandez-Campos, J.S. Marronb, G. Samorodnitsky, F.D. Smith: Variable heavy tails in Internet traffic, Performance Evaluation 2004 Comparison NGMN and Hernandez-Campos traffic model • NGMN model configuration 1 • 40% HTTP users 0,8 TR 36.814 0,7 in average 15kbit/s per user • 20% FTP users one type of objects reading time as in TR 36.814 in average 91kbit/s per user • 40% Video users 0,5 0,4 0,3 0,2 0,1 object size [bytes] 1,E+10 4,E+09 1,E+09 5,E+08 2,E+08 6,E+07 2,E+07 9,E+06 3,E+06 1,E+06 4,E+05 2,E+05 6,E+04 2,E+04 8,E+03 3,E+03 1,E+03 4,E+02 1,E+02 • Hernandez-Campos model 5,E+01 0 64kbit/s fixed bit rate per user 2,E+01 • Hernandez-Campos 0,6 7,E+00 • NGMN FTP 3,E+00 • NGMN HTTP 1,E+00 • parsing time and reading time according to weighted CDF • 0,9 • Based on Internet data traffic measured at the Campus of the University of North Carolina and is expected to reflect to a certain extend the today’s wireless traffic produced from increasing use of Smartphones and tablets • Object size distribution according to a mix of 3 log-normal distributions (implicitly includes all traffic types) • Load configurable via negative exponential IAT [NGMN] 3GPP TR 36.814 V9.0.0 (2010-03), Technical Report 3rd Generation Partnership Project; Technical Specification Group Radio Access Network; Evolved Universal Terrestrial Radio Access (E-UTRA); Further advancements for E-UTRA physical layer aspects (Release 9) [Hernandez-Campos] F. Hernandez-Campos, J.S. Marronb, G. Samorodnitsky, F.D. Smith: Variable heavy tails in Internet traffic, Performance Evaluation 2004 (University of North Carolina) Pooling Gains Resulting from User Load Distribution with Traffic Model Hernandez-Campos • Quantiles for 90%, 95%, 99% and 99.9% of the system time at its capacity limit Expected improvement up to a factor of 3 of the resource utilizations by the cloud approach between the uniform to nonuniform user distribution cases 100% 90% 99,9%tile 99%tile 95%tile 90%tile 80% used resources • 100% hotspot users: • less than 20% resource utilization for 90% of the time and • less than 38% for 99.9% of the time • 100% uniformly placed users: • less than 65% resource utilization for 90% of the time and • less than 91% resource utilization for 99.9% of the time 70% 60% 50% 40% 30% 20% 10% 0% 0 0,2 0,4 0,6 0,8 fraction of uniformly placed users 1 Resource Usage Trace for 50 % Uniformly Placed Users – Sum Over All Cells zoom zoom Hernandez-Campos Comparison of Resource usage for 50 % Uniformly Placed Users – Sum Over All Cells NGMN Hernandez-Campos Comparison Resource Usage for 100 % Uniformly Placed Users – Sum Over All Cells zoom zoom zoom Resource Usage Modeling: eNodeB Signal Flow S1 data rate header compression/decompression security functions handover support functions discard of user plane data due to timeout PDCP Buffer RLC buffer concatenation and/or segmentation of packets reordering of received HARQ packets RLC Scheduler MAC MAC-PHY rate FEC/FEC-1 QAM and precoding/receive processing 3/12/2013 Cell processing PHY2 rate User processing Multi-user scheduling, HARQ, ranging PHYuser * PHYcell *can be modeled with Desset 23 Processing Resource Modeling • Reference case: single-antenna, 64-QAM, rate-1 encoding and a load of 100% [Desset] • Scaling factors are exponents relating each contributor in the model to each parameter of the model Scaling exponent for code rate parameter C: s2 Scaling exponent for antenna parameter A: s3 Scaling exponent for layers parameter (streams) L: s4 Scaling exponent for Resource Block parameter R: s5 GOPS* in reference scenario for DL GOPS* in reference scenario for UL Scaling exponent for modulation parameter M: s1 FD (linear) 30 60 0 0 1 0 1 FD (non-linear) 10 20 0 0 DL: 2 UL: 3 0 1 FEC 20 120 1 1 0 1 1 Operation • Total Processing effort for user processing for the DL case Ptotal, DL PS1_ ter min ation, DL PPDCP _ ROHC , DL PPDCP _ ciphering, DL PRLC _ MAC , DL PPHYuser, DL PScheduler, DL with PPHYuser, DL PFD,lin , DL PS 1_ ter min ation, DL PPDCP _ ROHC , DL PPDCP _ ciphering, DL PFD,non lin , DL PFEC, DL 30 Aact / ref Ract / ref f S1 _ data_rate PRLC _ MAC , DL f S1 _ data_rate,type _ of _ HC PScheduler, DL 2 10 Aact / ref Ract / ref 20 M act / ref Cact / ref Lact / ref Ract / ref f S1 _ data _ rate, RLC mod e f resource _ allocation , # radiobearers f S1 _ data _ rate, type _ of _ sec *[Desset] C. Desset, B. Debaillie, V. Giannini, A. Fehske, G. Auer, H. Holtkamp, W. Wajda, D. Sabella, F. Richter, M., J. Gonzalez, H. Klessig, I. Gódor, M. Olsson, M. Ali Imran, A. Ambrosy, O. Blume, Flexible power modeling of LTE base stations, EARTH project 24 COPYRIGHT © 2011 ALCATEL-LUCENT. ALL RIGHTS RESERVED. ALCATEL-LUCENT — INTERNAL PROPRIETARY — USE PURSUANT TO COMPANY INSTRUCTION Comparison Resource Block Usage and of Processing Effort per Cell and per User Load 40% Even with good SINR, higher MCSs are selected rarely Comparison Resource Block Usage and of Processing Effort per Cell and per User Load 80% • At high load, interference increases and users with high SINR are rare • With high system load, higher MCSs become more rare • High processing load does occur, but is uncommon even if total system load is high. 26 COPYRIGHT © 2011 ALCATEL-LUCENT. ALL RIGHTS RESERVED. ALCATEL-LUCENT — INTERNAL PROPRIETARY — USE PURSUANT TO COMPANY INSTRUCTION Important Inputs from the Evaluation to the Resource Pooling Algorithms • Processing load and air interface load information is not sufficient • Introduce QoE metrics as additional criteria • GBR bearer: verify that the bit rate is guaranteed • non-GBR bearer: contribution of the RAN to the E2E latency requirement needs to be fulfilled • If there is some margin left for GBR bit rate or non-GBR latency, additional bearers can be admitted by taking into account the quality requirements of the other users • Decision is taken based on the air interface bandwidth and the processing capacity 27 COPYRIGHT © 2011 ALCATEL-LUCENT. ALL RIGHTS RESERVED. ALCATEL-LUCENT — INTERNAL PROPRIETARY — USE PURSUANT TO COMPANY INSTRUCTION Summary, Conclusion • Objectives and Requirements of C-RAN discussed • Mobile Cloud Architecture and Baseband Pool Functional Architecture presented • Several Processing split options incl. data rates are analyzed in pros and cons with a conclusion: the Interface PHY II or III for DL and for UL is a possible choice. -- 1*10Gb/s sufficient to drive a Multistandard radio site (40MHz LTE and 3*5MHz UMTS) • Processing split approaches and Virtualization Concepts for SOC BBUs and GPP configurations are presented with an approach for a MSS-BBU HW Architecture • Multi-cell System Simulation results: • significant pooling gains (20-80%) depending on spatial distribution of users and traffic model (implementation losses not taken into account) • different traffic models: less resource utilization variance with NGMN Model • Processing load and air interface load information is not sufficient • Introduce QoE metrics as additional criteria •Next steps: •Include HetNets •Extend processing resource model •Define adjustable measurements windows 28 COPYRIGHT © 2011 ALCATEL-LUCENT. ALL RIGHTS RESERVED. ALCATEL-LUCENT — INTERNAL PROPRIETARY — USE PURSUANT TO COMPANY INSTRUCTION
© Copyright 2026 Paperzz