Next Generation EPON- based Access Network Architecture

Access Network Link between the customer premises and the first point of connection to the network infrastructure—a point of presence (PoP) or central office (CO). Customer Premise Access Link Merto PoP / CO

Ethernet in the Last Mile

Access Bandwidth

Optical Access

What is Passive Optical Network ► Passive Optical Networks (PON) are point-to-multipoint optical networks with no active elements in the signals’ path from source to destination. ► Advantages of PON  PON allows longer distances between CO and customer: 20 km for PON vs. 5.5 km for DSL  PON provides higher bandwidth.  Allows downstream continuous broadcasting (video).  Eliminates electronic devices in the middle of the network.  Allows easy upgrades to higher bit rates or additional wavelengths.

Basic Architecture of PON

EPON Downstream

EPON Upstream

EPON Configuration

EPON Performance ► E ► EPON Media Access Control (MAC) uses Ethernet framing and line coding. ► ► Downstream channel uses true broadcast. ► ► Packets extracted by the MAC addresses. ► ► Not different from any shared-medium Ethernet LAN. ► ► Upstream transmission uses multiple access. ► ► Which multiple access scheme? (Problem)

Multiple Access Schemes

Statistical TDMA ► Time synchronization among ONUs cannot be easily achieved:  Who drives the clock?  How do we achieve synchronization? ► Ethernet in the first mile task force (IEEE 802.3ah) recommends Multipoint Control Protocol (MPCP).  Work is still in progress.  MPCP is not concerned with a particular bandwidth- allocation scheme.  MPCP supports mechanism that can facilitate various implementation of bandwidth allocation algorithms.

Timing Issues ► ► Ranging - RTT Measurement 1. OLT sends GATE at absolute T1 2. ONU receives GATE at T2, and resets local counter to show T1 3. ONU sends REPORT at time T3, showing timestamp T4 4. OLT receives REPORT at absoluteT5 RTT = T2-T1+T5-T3 RTT= T5-T4 T3-T2 = T4-T1

Multipoint Control Protocol (MPCP) Operation ► This protocol relies on two Ethernet messages: GATE and REPORT.  (Additionally MPCP defines REGISTER REQUEST, REGISTER, and REGISTER ACK messages used for an ONU’s registration.) ► A GATE message is sent from the OLT to an ONU.  It is used to assign a transmission timeslot. ► A REPORT message is used by an ONU to convey its local conditions (such as buffer occupancy, and the like) to the OLT to help the OLT make intelligent allocation decisions. ► Both GATE and REPORT messages are MAC (media access control) control frames (type 88-08) and are processed by the MAC control sublayer.

Statistical multiplexing ► ► Burst time and size are hard to predict. ► ► Must use schemes with feedback (like polling). ► ► Hub polling would work, but walk times are very large. ► ► Roll-call polling also works, but it requires ONUs to listen to each other.   PON should be deployed as a broadcasting star or passive ring (too restrictive). ► ► Proposed IEEE EFM standard solution: Interleave polling routines in time.

Interleaved polling scheme

Advantages of Interleaved Polling Scheme ► Bandwidth utilization.  If only one ONU is active, it can use up to 600 Mbps (with 5 μs guard band). ► Lower delay.  Delay is bounded by RTT, not frame time. Under maximum load behaves like TDMA system. ► No ONU’s synchronization necessary.  ONU sends data immediately on receiving (processing) the control message (Grant). No centralized framing necessary. ► All “smarts” are in OLT.  OLT may use various scheduling algorithms based on SLA, type of traffic, etc. ► Fast detection of disconnected ONU.  Disconnected ONU “consumes” only ~0.0005% of PON bandwidth.

Ethernet TCP/IP Frame 100Base CU Burst: Byte Frames per Burst

DBA Scheme  This algorithm is cycle-based, where a cycle is defined as the time that elapses between two executions of the scheduling algorithm.  The ONU will be granted the requested number of bytes, but no more than a given predetermined maximum W MAX (maximum transmission window). If Req i is the requested bandwidth of ONU i and Grant i is the granted bandwidth, Grant i is then equal to

Class-of-Service Considerations ► Performance in EPON can be characterized by several parameters:  bandwidth  packet delay (latency), delay variation, jitter  packet-loss ratio ► Quality of service (QoS) refers a networks’ to ability to provide bounds on some or all these parameters on a per- connection (flow, session) basis. ► Not all networks, however, can maintain per-connection state or even identify connections. ► To support diverse application requirements, networks separate all the traffic into a limited number of classes and provide differentiated service for each class. ► Such networks are said to maintain classes of service (CoS).

Overview of IEEE 802.1D Support for Classes of Service 1. Network control. Characterized by a “must get there” requirement to maintain and support the network infrastructure. 2. Voice. Characterized by less than 10-ms delay, and hence maximum jitter [oneway transmission through the local-area-network (LAN) infrastructure of a single campus]. 3. Video. Characterized by less than 100-ms delay. 4. Controlled load. Important business applications subject to some form of “admission control,” be that preplanning of the network requirement at one extreme to bandwidth reservation per flow at the time the flow is started at the other. 5. Excellent effort. Or “CEO’s best effort,” the best-effort-type services that an information services organization would deliver to its most important customers. 6. Best effort. LAN traffic as we know it today. 7. Background. Bulk transfers and other activities that are permitted on the network but that should not affect the use of the network by other users and applications.

Dynamic Bandwidth Allocation

Timeslot utilization is less than 100% ► ► Packets cannot be fragmented. ► ► If the next packet to be transmitted is larger than the remainder of timeslot, the packet will wait for the next timeslot => the timeslot will be transmitted with an unused remainder at the end.

Why timeslot adjustment won’t work ► ► Why timeslot adjustment won’t work ► ► Linear increase in offered load requires exponential increase in timeslot size. ► ► Increased timeslot size will increase timeslot period => will increase packet delay. ► ► Timeslot adjustment should be based on traffic load. ► ► However, due to burstiness of traffic at every timescale, no load prediction is possible based on previous load.

Drawbacks of OLT based DBA ► OLT-ONU is 20km and a control messages (REQUEST and GRANT) consumes significant portion of the valuable upstream bandwidth. ► ONU’s traffic changes dynamically and very bursty in nature thus most recent buffer status is not at hand when OLT makes DBA allocation. ► CoS cannot be truly support by centralized DBA decision as OLT relies on inter-ONU scheduling for optimal solution and hence fails to take into account critical QoS parameters while arbitrating between ONUs.

OLT ONU  Control Plane:  1310nm channel  Data Plane:  Upstream: 1310nm channel  Downstream: 1550nm channel 1550nm 1310nm Redirected 1310nm signal Splitter/ Combiner Proposed New PON Architecture (In-band Signaling)

3xN S/C ONU OLT 3xN S/C ONU OLT 3xN S/C ONU OLT 3xN S/C ONU OLT Control Data a) First Phase b) Second Phase c) Third Phase 3xN S/C ONU OLT 3xN S/C ONU OLT a) First Phase b) Second Phase c) Third Phase Algorithm (DBA) Individual ONU update messages Combined ONU update messages Individual ONU data messages Combined ONU data messages [Time] Combining of ONU update messagesCombining of ONU data messages Combined ONU data messages Combined ONU update messages

Distributed DBA for EPON: In-band Control Plane ► Using (Splitter/Combiner) we reflect 1310nm upstream bound signal. ► We use REQUEST Control frames to update all ONU’s of the current ONUs’ buffer info. ► After receiving all updates from all ONUs (max. 64), each ONU independently run DBA and arrive at one unique timeslot allocation per ONU. ► A copy of the REQUEST also propagates to OLT and it also can run the same DBA to know which ONU is transmitting when. ► CoS could be easily factored into the DBA decision.

Distributed DBA for EPON: In-band Control Plane (Cont.) ► A portion of the upstream bandwidth is consumed to establish the control plane, however it is very small (less than 5%). ► Time synchronization among ONUs is an issue:  Fixed downstream frame sizes could be used to derive time synchronization.  The average radius from the Splitter/Coupler to ONUs is less than 1km and we propose to have a fixed distance of 1 km to avoid time delay issues. ► The proposed cycle time (window size) is 2ms  Optimized cycle time would be investigated under various traffic load and QoS requirements.

OLT ONU Splitter/ Combiner 1550nm 1310nm  Control Plane:  Fixed Wireless LAN  Data Plane:  Ethernet Passive Optical Network Proposed New PON Architecture (Out-of-band Signaling)

Data Data Plane i i+1 Control Plane Control Distributed DBA for EPON: Out-of-band Control Plane ► Since ONUs are with in less than 2km diameter, we can use fixed wireless to establish the control plane. ► Control information from the ith window is used to run DBA for timeslot allocation per ONU. ► Out-of-band signaling relieves the upstream channel to be fully utilized for data traffic only.

Thesis Proposal ► To develop and implement a fully distributed EPON-based dynamic bandwidth allocation algorithm. ► The work will be carried out in two stages:  Simulation studies using OPNET and other tools.  Physical implementation of DBA in the lab test bed.  Simulation data will be compared to the empirical data obtained from the lab experiments. ► The proposed Next Generation EPON-based Architecture will unleash the Access bandwidth bottleneck and support total packed-based QoS guaranteed new applications.

3X3 Splitter/ Combiner Isolator Workstation1 (ONU) Workstation 3 (ONU) Workstation2 (ONU) Server (OLT) Wireless Access Card GigE Card SM Fiber (500 m) SM Fiber (20 Km) Testbed SETUP