Evaluating the performance of Network Equipment. Presenter: Tommy Cook, CEO Calnex Solutions Ltd

Transcription

1 Evaluating the performance of Network Equipment Presenter: Tommy Cook, CEO Calnex Solutions Ltd

2 Presentation overview Proving performance of; EEC Synchronous Ethernet Devices. 1588v2 Boundary s. 1588v2 Transparent s. 1588v2 Ordinary s. Networks carrying 1588v2. COMPANY CONFIDENTIAL 2

3 EEC Synchronous Ethernet devices COMPANY CONFIDENTIAL

4 Sync-E Wander to ITU-T G ) Wander generation 2) Wander tolerance Ref SyncE (Wander-free) Measure Wander EEC Ref SyncE + wander EEC Maintain clock within performance limits. Not causing any alarms. Not causing the clock to switch reference. Not causing the clock to go into holdover. Check ESMC 3) Wander transfer 4) Phase transient response Ref SyncE + wander Measure Wander EEC Ref SyncE (Wander-free) Measure Phase transient EEC Switching between different synchronization signals results in phase transients at the EEC output COMPANY CONFIDENTIAL 4

5 Sync-E Jitter to ITU-T G ) Jitter generation 2) Jitter tolerance Ref SyncE (Jitter-free) Measure Jitter EEC Ref SyncE + Jitter EEC No dropped packets indicating bit errors being introduced. Not causing any alarms. Not causing the clock to switch reference. Measure Dropped Packets. Monitor ESMC Status. COMPANY CONFIDENTIAL 5

6 ESMC to ITU-T G.8264 Change ESMC to PRC 1.Change the ESMC status being received by EEC. 2.Verify ESMC response, and prove EEC also switches clock source by monitoring Wander. Change ESMC to DNU Wander on Output of DUT Locked to Local Ref. Locked to Ref. Locked to Ref. Ref Change ESMC Local Ref ESMC on Output of DUT EEC QL-PRC Monitor Wander & ESMC Response QL-EEC1 COMPANY CONFIDENTIAL 6 06

7 COMPANY CONFIDENTIAL 1588v2 Devices

8 Performance objectives of 1588v2 network MKR-1:x= , y= MKR-2:x= , y= Delta: x= , y= x= y= t-ms t1 Master Sync Slave Time &/or Frequency Output t2 Time Interval Error (TIE) vs Nominal (seconds) T (a) Offset=0.002 ppm Performance is specified on the Time &/or Frequency output from the Slave clock. t-sm t4 Follow_Up Delay_Req Delay_Resp Propagation Delay Message Exchange t3 Frequency Output must comply with the relavent ITU-T interface specification, (MTIE & TDEV specification); G.823/4 Traffic Interface Masks. G.823/4 Sync Interface Masks. Time output must comply with end application requirements. 1pps output Time error in µsec COMPANY CONFIDENTIAL 8

9 Designing your network On-path support: Ethernet switch has either Boundary (BC) or Transparent (TC) functionality to control the impact of the network on the 1588v2 messages, easing the challenge of recovering the Time and Timing from the 1588v2 messages. Sw (1) Sw (1) Sw (2) BC (a) Sw (3) No On-path support TC (b) Partial On-path support Sw (4) Sw (2) Sw (n) Sw (m) Slave Slave PRC GM BC (1) TC (a) BC (2) TC (a) TC (b) TC (p) BC (q) Slave Full On-path support COMPANY CONFIDENTIAL 9

10 COMPANY CONFIDENTIAL Boundary s

11 Boundary Q1 Slave Master Q2 Boundary s reduce PDV accumulation by; Terminates the PTP flow and recovers the reference timing. Generate a new PTP flow using the local time reference, (which is locked to the recovered time). No direct transfer of PDV from input to output. Qn Boundary is in effect a back-to-back Slave+Master. COMPANY CONFIDENTIAL 11

12 TIE TIE Sources of PDV from Boundary Q1 Slave Master Q2 Potential Sources of PDV; a) Wander; Each BC recovers the clock and re-generates a new timing signal. This can lead to the introduction of low frequency clock wander, Chains of BCs can lead to the accumulation of low-frequency clock wander. Time Regenerated vs. GrandMaster Qn Output PDV Time b) High-freq. PDV from internal BC packet management; PDV from Output Buffer Queue of BC; 12 µsec for 1514 Byte packet, (G.8261). 525 µsec for 64K Jumbo Byte packet PDV from other internal queues. Affected by other High Priority Traffic? COMPANY CONFIDENTIAL 12

13 Performance specification of a BC Draft ITU-T G will specify the performance of a BC. Four sections have been proposed; 6. Noise Generation 7. Noise Tolerance 8. Noise Transfer 9. Phase Transient and Holdover Response The approach being taken in G is following the well established methods of specifying the performance of node clocks (e.g. in G.8262 for SyncE, etc.) The impact of congestion traffic on the output PDV may require an additional test; X. Impact of congestion traffic COMPANY CONFIDENTIAL 13

14 TIE TIE TIE TIE TIE Boundary Test Plan to G Measure PDV Wander 1) Noise generation 2) Noise tolerance Inject PDV Wander Master `1` Wander free BC PDV Meter Time Slave `1` Master `1` PDV Meter BC Time Slave `1` Check for alarms &/or BC going into hold-over 3) Noise transfer 4) Phase transient response Inject PDV Wander Measure o/p vrs i/p Wander Switching between synchronization signals. Measure Phase Transient Time Time Time Master `1` PDV Meter BC PDV Meter Slave `1` Master `1` BC PDV Meter Slave `1` COMPANY CONFIDENTIAL 14

15 Boundary Test Plan: Impact of congestion traffic. ~ Master `1` BC PDV Meter Slave `1` Development Test Procedure to characterise actual performance Traffic Generator Congestion traffic e.g. conforming to G.8261 VI.2.2 Network Traffic Model 2 1. Measure PDV induced by policing and buffer mechanisms; 1. Vary traffic packet size. 2. Vary traffic priority. 3. Vary traffic utilisation. 2. Test in 1-Step and 2-Step modes. COMPANY CONFIDENTIAL 15

16 COMPANY CONFIDENTIAL Transparent s

17 Transparent Transparent s reduce PDV by; Calculating the time a PTP packet resides in the TC device (in nsec) and insert the value into the correctionfield. By using the correctionfield, the Slave or terminating BC can effectively remove the PDV introduced by the TC. Q1 Q2 Qn Packet Delay in TC Device inserted into correctionfield at output of Transparent device COMPANY CONFIDENTIAL 17

18 Accuracy of the Correctionfield value: Does it reflect the actual delay experienced by the Sync & Del_Req messages? Theoretical model: CorrectionField precisely reflects the delay through the equipment Ideal case zero net PDV Q1 Q2 Qn Potential Sources of inaccuracy; 1. PDV introduced dependent on vendor s implementation; Impact of Output Buffer Queue not removed. 12 µsec for 1514 Byte packet, (G.8261). 525 µsec for 64K Jumbo Byte packet Varying sized Packets not corrected to same accuracy. 1-step and 2-step processes produce different behaviour. Does it make a difference if High Priority Traffic is present? 2. Is the TC correction effective in both directions? COMPANY CONFIDENTIAL 18

19 ~ Transparent Test Plan Development Test Procedure to characterise actual performance Master `1` TC Traffic Generator PDV Meter Congestion traffic e.g.conforming to G.8261 VI.2.2 Network Traffic Model 2 TC Slave `1` IEEE std C PTP in Power Systems Applications. Annex A: TC TimeInaccuracy 50nsec. 1. Measure impact of correctionfield on Sync PDV. 1. Vary traffic packet size. 2. Vary traffic priority. 3. Vary traffic utilisation. 2. Repeat for Del_req PDV. 3. Test in 1-Step and 2-Step modes. Run G.8261 Test case scripts and measure PDV with and without use of CorrectionField. For high accuracy measurement, measure PDV at input & output concurrently and calculate difference including the correctionfield. COMPANY CONFIDENTIAL 19

20 COMPANY CONFIDENTIAL Ordinary s

21 Prove performance in presence of congestion PDV: Master Slave MKR-1:x= , y= MKR-2:x= , y= Delta: x= , y= x= y= Time Interval Error (TIE) vs Nominal (seconds) Congestion in both directions will impact clock recovery. Master T (a) Offset=0.002 ppm Slave Time &/or Frequency Output G.8261 Appendix VI Test Cases are the most widely used approach to verifying operation. PDV: Slave Master ~ MKR-1:x= , y= MKR-2:x= , y= Delta: x= , y= x= y= Synchronisation Source Time Interval Error (TIE) vs Nominal (seconds) Offset=0.002 ppm T (a) COMPANY CONFIDENTIAL 21

22 COMPANY CONFIDENTIAL Networks

23 PDV Metrics ITU-T G.8260: Appendix under development that defines a number of metrics that may be used for analysis of PDV. WD8260ND_G8260, WP3/15; ITU-T Recommendation G.8260 Latest Draft, York Sept. 11 Metrics defined include; MATIE, minmatie - Maximum Average Time Interval Error MAFE, minmafe - Maximum Average Frequency Error mintdev, PercentileTDEV, BandTDEV (TDEV Time Deviation) ClusterTDEV Floor delay packet population. pktfilteredtie, pktfilteredmtie, pktfilteredtdev, pktfilteredffo COMPANY CONFIDENTIAL 23

24 PDV Metrics Objective: Extract and identify the key characteristics of the PDV that impact the ability of the Slave to lock to the Master clock. Block Diagram from ITU-T Standards Local reference PEC-S Packet Timing Signal Packet Selection Time Scale Comparator Low Pass filter Oscillator Output Interface: Ethernet (packets) Local Time scale Interface: E1 or T1 Metric: pktfilteredmtie, pktfilteredtdev Metric: MTIE, TDEV COMPANY CONFIDENTIAL 24

25 PDV pktfiltered Metrics Packet Time Error Sequence Selected-Packet Time Error Sequence Filtered-Packet Time Error Sequence x(t) x (t) y(t) Packet Selection Bandwidth Filtering Traditional Metric G.8260 Figure X - Packet selection & Filtering Flow time time time MTIE PDV Stage 1: Packet Selection Stage 2: Low Pass Filter Stage 3: MTIE/TDEV Analysis Select Floor Packets, lowest X%. Band, X-Y% Cluster Window Size. Window step size. 1 st or 2 nd Order LPF. Perform Traditional Standards-based algorithms e.g. MTIE, TDEV, etc. COMPANY CONFIDENTIAL 25

26 Summary of Evaluation Plan SyncE 1588v2 BC TC Ordinary s Networks Standards in force. G.8262 Wander Jitter G.8264 ESMC behaviour G Standard under development. Noise Generation Noise Tolerance Noise Transfer Phase Transient & Holdover; Consider behaviour during Traffic Congestion. Prove accuracy of CorrectionField. Characterise specific TC behaviour. IEEE Std C PTP in Power Systems Applications Profile specify 50nsec. Character using G.8261 Appendix VI test cases. A number of metrics under development in Draft Appendix in G A number of metrics under development in Draft Appendix in G pktfilteredmtie, pktfilteredtdev MAFE, mintdev, etc COMPANY CONFIDENTIAL 26

27 Tommy Cook +44 (0) Calnex Paragon Sync IEEE 1588v2 CES Sync-E OAM COMPANY CONFIDENTIAL 27