Power saving in (X)GPONs

1. Joint ITU/IEEE Workshop on Ethernet - Emerging Applications and Technologies(Geneva, Switzerland, 22 September2012) Power saving in (X)GPONs Frank Effenberger Rapporteur Q2/15 VP Access R&D, Futurewei

2. Introduction • Power saving in the ITU • Supplement G.45 • Actual power consumption

3. Initial activities • First contributions were made by semiconductor vendors, to consider signaling methods • The Q2/15 group thought these were premature • The requirements for power saving were not clear • The impact on existing systems was not clear • There was a concern it would degrade the user experience • It was agreed that a survey would be made of the operators to learn their requirements

4. Power saving survey #1 • Survey gathered information on • Basic power supply designs • Who pays, Who changes the battery? • Overall requirements and interest in power saving • The most telling result was: Which is a higher priority, service availability or power savings? • The answer was overwhelmingly “Service quality is much more important that saving power” • Apparently, the “Green Revolution” had not yet happened

5. Power saving surveys #2 and #3 • Survey #2 focused on the case of power failure • What UNI’s are in common use? • How do they get powered down if the main power fails? • Who can control this powering down process? • Survey #3 focused on regulations about lifeline • Are their regulations that force the maintenance of service during a power failure? • How do those regulations vary from service to service (POTS vs. video vs. Internet)?

6. Outcome of the surveys • It was clear that there was interest in power saving • For the most part, power saving was seen as a way to lengthen the life of the battery during an outage • Operators were not willing to compromise much in the name of power saving • Can’t cost any more • Services can’t be effected • This set the stage for the G.sup45 document

7. Introduction • Power saving in the ITU • Supplement G.sup45 • Access power consumption

8. Outline of G.sup45 Requirements Classification of techniques Signaling of ONU operations Comparative analysis Conclusions

9. Requirements • Surveys were used as a primary requirement • Power saving mode should be triggered by power failure, and NOT low traffic or unused ports • EU CoC • The current state of these power targets was reviewed • The “low power” mode was noted to be only half of full power mode • There are two main requirements • To maintain service during a power failure • To save power at all times • No operator consensus on the balance between these two requirements

10. Classification of techniques

11. Power Shedding • When triggered, the ONU turns off the UNI’s • Turn off in this context means fully powering down the circuit (not just deactivating service) • Each UNI type can have a different shut-down period (e.g., video can turn off after 30 minutes, but POTS is maintained for several hours) • When to trigger is a question • During power failure, some UNI’s can be turned off safely (e.g., video, because the TV sets will not have power in most cases) • During normal times, it is difficult to judge if UNI is busy • This is the least service effecting method • ONU maintains contact with OLT at all times

12. Dozing • When triggered, the ONU should stop transmitting in the upstream, even if it is given BW allocations • This allows transmitter circuitry to go into low power mode • Lower than normal “off” in between bursts • May take longer to recover (10’s of milliseconds) • Trigger would be the inactivity of the ONU • Difficulty is that data services (and VoIP) tend to “chatter” all the time • Downstream receiver and signal chain remains on • ONU can be signaled by the OLT • Incoming calls can be received without delay • This impacts services slightly • Outgoing communications might suffer a delay, as normal bidirectional communication is reestablished

13. Fast Sleep • When triggered, the ONU shuts off entire PON interface for a short period of time • ONU periodically wakes up to see if OLT has anything to say • During the shutdown, the ONU could conceivably have nearly zero power drain (only the wake-up timer would be powered) • Key issue is how fast can you wake up the optics • Normal transceiver designs are not optimized to turn on fast • However, proper optimization could get times down to ~1 ms • Side note: Recent result have shown circa 60% reductions • This method can have relatively low service impact • ONU maintains contact with OLT (albeit transiently) • Interactions with higher layer protocols must be considered

14. Deep Sleep • When triggered, the ONU completely shuts off • Services are definitely impacted, no apologies for that • Power drain is zero, or nearly so • Challenge: How to wake up? • Snow White method: A prince (the user) kisses the deep sleeper (presses a button on the ONU) • Rip Van Winkle method: Deep sleeper wakes up after a preset time, and sees if anything has changed • This method only appropriate for long outages • It seems that the usual obligations are excused if power is out for a long time, and users and regulators understand that

15. Signaling of ONU operations • Dying gasp: Enhancing the existing message • For G-PON, not accepted, because it changes the TC-layer • PLOAM-based: Signaling for fast sleep method • For G-PON, not accepted , because it changes the TC-layer • OMCI-based: Configuration of power features • For G-PON, OMCI additions have been made • Extended Power shedding: Detailed control • For G-PON, fine-grain control of shedding has be standardized • Implicit signaling: OLT suppresses alarms • No standards impact, so OLT vendors are free to implement • Security aspect: Impostor attack • When the ONU is asleep, impostor can more easily jump in

16. Comparative analysis • Model of ONU power consumption is given, and used to evaluate the savings for each type power saving • This model is only an example, based on a particular ONU design and circuit power values (these change over time) • Key findings of this evaluation • Power shedding accomplishes a lot (70%) of power saving • The other methods have increasing implementation difficulty and declining efficacy of power saving

17. Conclusions of G.sup45 • Power saving is an important topic • Main object is to improve handling of power failures • Recommendations to improve power usage • Continuous improvement of design (ASIC, optics, power conv. Etc.) • Power shedding should be supported and activated • Dozing can be implemented with little cost • “Aggressive” sleeping modes are of lowest priority • Final note: G-PON saves power in ICT field and other industries, so some credit should be given for that

18. Introduction • Power saving in the ITU • Supplement G.45 • Access power consumption

19. System architecture of a VDSL system DSLAM CPE LT SW WAN LT LT CPE SW WAN LT LT CPE Typical VDSL linecard consumption today is 2W per line (i.e., per user) Typical VDSL HG CPE consumption is 10W per user

20. System architecture of G-PON system OLT ONU LT SW WAN LT LT ONU SW WAN LT LT ONU Typical OLT linecard consumption today is 7W per PON port (@ 28 users/PON = 250mW/user!) Typical GPON HG CPE consumption is 10W per user

21. Central office / Node “power crunch” • Central office dissipation is dictated by NEBS • Typical US number: 2000W per bay, 3 racks per bay • Typical DSLAM has 16x24 lines = 768W per rack • This barely fits in the 2000W number • Typical OLT has 16x8 PONs = 896 W per rack • Have to leave 1 rack-space empty! • Is PON hitting the “crunch”? NO! • One OLT serves 3584 users, while a DSLAM serves only 384 users • We need 9 times fewer OLTs than DSLAMs

22. Trend in broadband access CO equipment • The power per chassis is increasing marginally • Perhaps a 30% increase generation-over-generation • The capability per chassis is increasing incredibly • Aggregate bandwidth increases 4~10x per generation • Users per chassis increased ~10x from copper to fiber • Total access power per user is already declining • Driven by the acceptance of fiber access • Power density is increasing • Suggests a rethinking of the CO power design guidelines • Perhaps even a redesign of the cooling method entirely • Diffused air cooling (typical in today’s CO) is inappropriate for intense point heat loads

23. The CPE power issue - Functional blocks Mem POTS Interf (x2) WAN interface MAC Ethernet Interf (x2) Typical Single family home gateway CPE consumption is 10W

24. Breakdown of ONU (VDSL is similar)

25. Observations on baseline consumption • Power consumption is reasonably balanced amongst functions – there is not one “bad actor” • The majority of power (60%) lies in functions that are not particularly related to PON • You find them in any access system • Many are legacy dictates (ringing a bell) • They are designed for reliability and performance • E.g. Power converters consuming 20% of the power… why? To handle the stress environment that Telco requirements give us • Flexible hardware (e.g., CP instead of ASIC) is used • The flexibility is a meta requirement of the ever changing market • But, this is never the most power efficient way to build equipment • If the true power cost of all the requirements was rationalized, just imagine what we might save!

26. Power saving technologies • Most natural path is intrinsic improvement • Current designs were not designed with power as a key requirement • Time to market, performance, and simplicity were always more important to the designer • Example: burst mode laser driver • The average ONU duty cycle is ~3% (32 ONUs per PON) • But, the typical laser driver consumes current 100% of the time • Why? Because it was easier that way • This is straightforward to fix • The designers only need to be guided that power consumption is an important goal that has value • This process is underway already!

27. “Always on” means “always polluting” • Recall the original telephone network • You only used power when off hook – very efficient, and natural behavior to the user • Data separated the “session” from user • Ideally, users interact with their computer, and the computer establishes the (logical) sessions automatically • User involvement in session control (dial up) was slow and painful • This quickly drove the “always on” model • Power consumption was not considered!

28. Sleep modes for access equipment ONU state diagram OLT state diagram Protocols for sleeping and dozing are standardized in the XG-PON system

29. Future possibilities • OLT power consumption could be reduced in future PON systems • OLT “shedding”: If a port is not used, it should be powered down • As deep into the card as possible • OLT “sleeping”: If a TWDM-PON is underused, reduce the active waves • ONUs would be concentrated onto fewer channels • This could improve the load-dynamic power consumption of the CO

30. Getting a good night’s sleep • Standardization is only the beginning • The hardware must be designed to use it • Optoelectronics must have fast turn-on/off • Logic devices must support the protocols • Switching must recognize that link is transient • The operators must be motivated to use it • Operators respond to competitors and users • In a choice between performance and power-saving, which wins? • Example: ADSL has power saving for some years now – almost never used

31. Conclusions Current access power consumption is trending in the right direction, considering the incredible BW improvements The CO-side solution is in our hands: deploy PON, and you cut your CO power by an order of magnitude The CPE-side is much larger problem Legacy interface requirements are an issue If we could only redesign POTS… If only Telco’s could agree on a service profile and stick to it… Power saving modes have good potential Changing “always on” into “always available” Already standardized – we just have to do it