Technology Evaluation for Tim e Sensitive gy Data Transport Report and status for subtask in JRA1 GN3 JRA1 Workshop, 20st-22nd of November, Copenhagen Henrik Wessing, Task leader
Tim e Sensitive Transport - Agenda • Motivation and objectives • Motivation and objectives – ”Is it worth establishing circuit based paths or do we survive with the packet based solutions?” • Technologies and layering • Technologies and layering • Delay verification through test equipment and OAM • Physical layer • OTN layer • OTN layer • MPLS-TP and PBT • Measurements • Conclusions and continued work • Conclusions and continued work
Low delay is a key property • Synchronising atomic clocks • Synchronising atomic clocks • Online music production (LOLA) – Low Latency audio and video conferencing – Delay < 35-40 ms (for network) y ( ) – Jitter < 3 ms – Characterised by many small packets Picture from presentation by Claudio Allochio, GARR • Telemedical surgery – Roundtrip delay < 150 ms – Gb/ s bandwidth due to no compression • Interconnection of data centres – NREN operating as network for DC – NREN operating as network for DC. • Banking – Stock exchange algoritms can utilise reduction of delays in msec and usec range – Not a clear NREN task! – According to The Telegraph: • ”… that a one millisecond advantage could be worth up to $100m a year to the bottom line of a large hedge fund.”
And m ore and m ore services em erges Delay tolerant Delay sensitive Remote backup ve or Sensitiv 3D streaming Bandwidth Grid computing Internet surfing with Thin Clients • Variety of services video Tele Health Internet surfing B Err Email • Different requirements Mobile gaming • End to end QoS in focus End to end QoS in focus Ultra HD TV Ultra HD TV r tolerant Stereoscopic TV ndwidth (U)HD Video Stream Video Conferencing • Delay key issues Video Streaming IP Telephony Ban Error Home monitoring Online Distributed Environments
Technologies investigated • Campus networks / Metro networks C t k / M t t k – Traditionally packet based (L3/ L2) depending on operational use • Core / NRENs (GEANT) – Packet or circuitbased P k t i itb d • Methodology – Layered approach from physical layer and up – Qualifying and quantifying delay parameters Service (IP) Layer 2 (ETH / MPLS-TP) L1 S b L1 Subwavelength layer (OTN - ODUk) l th l (OTN ODUk) L0 DWDM layer
Measuring delay – External test or OAM • Delay measurements using external test equipment • Delay measurements using external test equipment • Delay measured between Maintenance association End Points – Different levels acoording to domain EF 35 From M
OAM – Delay and loss statistics • • Metro Ethernet Forum specifies delay statistics (MEF 10 2) Metro Ethernet Forum specifies delay statistics (MEF 10.2) • Frame delay (range/ mean/ jitter) - histograms • Frame Loss Ratio - counters • Availability • • One way FD utilising timestamps and sequence numbers (MEF 35 IA) One way FD utilising timestamps and sequence numbers (MEF 35 IA) – Real data or as added synthetic frames – Synchroneous clocks required – (or estimated from two way FD) • Performance Monitoring solutions – Choice depending on single or dual ended Choice depending on single or dual ended – Synthetic frames must match real frames • Communication messages – Delay Measurements Message (DMM)/ Delay Measurement Response (DMR) • R Results stored in bins for each measurement interval lt t d i bi f h t i t l PM MEG Measurement PM Function(s) Mandatory Solution Type(s) Technique for Loss or Optional point ‐ to ‐ point Synthetic Single ‐ Ended Delay PM ‐ 1 Mandatory multipoint Testing Single ‐ Ended Synthetic Loss point ‐ to ‐ point PM ‐ 2 n/a Dual ‐ Ended Delay Optional multipoint Counting g PM ‐ 3 PM 3 point ‐ to ‐ point i t t i t Si Single ‐ Ended Service Loss l E d d S i L O ti Optional l Service Frames
Dedicated test equipm ent • Agilent (now Ixia) N2X test solution • Main advantage: You define which packet sizes and load to test with • Main advantage: You define which packet sizes and load to test with • RFC2544 compliant • Measuring loss, signal power, delay, jitter etc. • From services to fibres • From services to fibres
Physical layer ( L0 ) • Physical layer (L0) • Physical layer (L0) – Propagation delay in fiber. Well defined. • 1 km approxim ately 5 usec – DCF adds delay. May use DCM. DCF adds delay May use DCM – Delay in modulation formats depending on technology in sub usec range • QPSK QAM OFDMA • QPSK,QAM, OFDMA … – Transponders/ Muxponders • ~ 5-10 usec For m edium to long range applications: Propagation is the only significant L0 delay
Optical Transport Netw ork ( OTN) – L1 • Physical Media Independent layer (L1) • Physical Media Independent layer (L1) - general general – Mapping from higher layer to L1 – Forward Error Correction • OTN OTN – Basically a technology to replace SDH/ SONET better suited for ethernet transport. – Timestamping for Delay Measurement Message and Delay Timestamping for Delay Measurement Message and Delay measurement reply (DMM/ DMR) – Vendor support: All major vendors support ODU switching
OTN ( I I ) – Measurem ent scenarios • Lab measurements using Alcatel Switches • Lab measurements using Alcatel Switches Loops in Delay contribution – Triangular setup triangle 1 1 GFP mapping and demapping • CPH - CPH -HAM Total fibre propagation delay • Loops in triangle to estimate delays from p g y mapping and FEC processing 2 x OTN switching latency • Expected delays 2 1 GFP mapping and demapping – usec range (OTN mapping and processing) • Results using external N2X tester and OAM Res lts sing e te nal N2X teste and OAM T t l fib Total fibre propagation delay ti d l information Triangle fibre propagation delay 5 x OTN switching latency 3 1 GFP mapping and demapping N2X tester OTN switch Total fibre propagation delay 2 x Triangle fibre propagation delay 8 x OTN switching latency OTN switch OTN switch OTN switch OTN switch
L2 – MPLS TP and PBT – Packet handling • • MPLS TP MPLS TP – Two flavours: MPLS-TP and T-MPLS – No signalling protocol – Legacy Ethernet can be encapsulated in PW transported over MPLS LSP – Delay contributions Delay contributions • Label processing • Store and forward or wire speed as no CRC FCS FCS • Lower priority packet may delay if in process Payload FCS = = Payload l d 802.1ad frame with or FCS without Payload • PBT FCS TPI D – Ethernet for transport purposes C-TAG I -TAG 32 bits Payload TPI D TPI D 16 bits 16 bits TPI D TPI D TPI D TPI D – VLAN -> Q-in-Q -> MAC-in-MAC VLAN Q i Q MAC i MAC Q-TAG S-TAG B-TAG 16 bits – Delay contributions 16 bits TPI D TPI D TPI D TPI D • MAC table lookup SA SA SA B-SA 48 bits • Store and forward due to CRC DA DA DA B-DA 48 bits – Dependent on packet length • Lower priority packet may delay if in process
PBT Lab scenarios Simple PBT Delay in 1, 2, 3, 4 and 5 switches Results next page p g PBT switch s tc PBT switch PBT switch N2X tester PBT switch PBT switch PBT over OTN Mapping between OTN and PBT Mapping between OTN and PBT - Crossconnect version - Terminate in PBT version
PBT m easurem ents – Sim ple PBT 140,00 64 bytes 120,00 512 bytes 100,00 1500 bytes 80,00 60,00 40,00 20,00 0,00 0 1 2 3 4 5 • Delay in usec depending on number of PBT switches • For Jumbo frames of 9000 appr 80 usec per node (400 usec for 5 nodes) • For Jumbo frames of 9000 appr. 80 usec per node (400 usec for 5 nodes) • Clear dependence on packet length • Processing max ethernet size packet equals 4-5 km transmission!! – For Jumbo: ~ 16 km transmission! For Jumbo: 16 km transmission!
Dependence on load p • All numbers in usec • Jumbo frames included only in text 25 • Vendor specific • Vendor specific 64 20 • 1Gbps 512 • Graph: Delay dependent on 1500 15 load Delay 10 Load 64 512 1500 9000 10 9,203 13,77 22,741 82,785 20 20 9 208 9,208 13 779 13,779 22 743 22,743 82 787 82,787 5 30 9,221 13,784 22,748 82,781 40 9,214 13,766 22,74 82,779 50 9,232 13,781 22,732 82,773 60 9,233 , 13,781 , 22,732 , 82,776 , 0 10 20 30 40 50 60 70 80 90 100 70 9,325 13,783 22,746 82,787 80 9,439 13,795 22,74 82,791 90 9,561 13,818 22,746 82,775 100 10,157 14,742 23,602 83,554
Recommend
More recommend
