The ONT makes use of the 802.1p-bit bridging concept – to enables the ubiquity of prioritybased services offered in PON networks. In upstream, the main idea is to allow the 802.1p market Ethernet packets to be forwarded to the right GEM Port based on user port, VLAN ID, VLAN priority, Ethernet Type, or DSCP value. The GEM Port is attached to a T-CONT, which is pulled at a certain frequency at the discretion of the OLT. In the downstream direction, the ONT terminates the GEM layer, extracts the Ethernet packets, and forwards them into the right downstream priority queue based on their 802.1p priorities. The p-bit is an integer value from 0 to 7 that is either encoded in the tag header of a priority-tagged and VLAN-tagged frame, or derived from the DSCP number in the IP header of an untagged frame. Where an untagged frame from a trusted source is received, the mapping is used to convert the DSCP number in the IP header of the frame to a p-bit value. The mapping of DSCP numbers to p-bits is configurable at the system level and at the UNI level.
1.1.1.
UPSTREAM ONT QOS ARCHITECTURE
Each Ethernet port on the ONT (or ONT UNI) is provided with eight queues. The ONT UNI port can support multiple traffic schedulers in the upstream direction, where each scheduler is allocated to an individual service on the UNI. In addition, the (relative) weights of the individual queues are configurable; Strict Priority scheduling is achieved by allocation a zero weight to the queues, whereas a value between 1 and 255 will trigger Weighted Round Robin (WRR) handling. Note that; 256 will put the queue out of service. In case of two SP queues on the same scheduler, the queue with the highest pbit value will get the highest priority to be served.
The mapping of the upstream traffic into the individual queues and schedulers depends on the UNI classification mode – as detailed in Error! Reference source not found..
PortIDx1 … PortIDx8 PortIDy1 … PortIDy8 T- CON T UN I Service model for T- CON T/ GEM Port a rra ngement queue1 T- CON T
HSI VoIP VoD
SP/ WRR
… queue8
ONT Interface
queue1
HSI
PortIDz1 … PortIDz8 T- CON T
SP/ WRR
… queue8
VoIP
ONT Interface
ONT
Figure 1: ONT upstream QoS architecture
1.1.2.
DOWNSTREAM ONT QOS ARCHITECTURE
In the downstream direction the ONT may be receiving more data than its egress ports can support. A substantial amount of data buffering is provided to absorb any data bursts until they can be transported towards the customer. In order to provide the right precedence to the highest priority traffic, the UNI supports a SP scheduling mechanism in the downstream direction.
PortIDx1 queue1 … PortIDx8 PortIDy1 … PortIDy8 UN I Service model for GEM Port arrangement
… queue8
Whereas the upstream QoS handling provides flexibility in terms of configuration of individual schedulers per service on the UNI, all the downstream traffic from all the service is scheduled through the same UNI SP scheduler. The mapping of the aggregate traffic of all the UNI services is based on pbits only, whereby the highest priority pbit (i.e. pbit 7) will be allocated to the highest priority queue (i.e. Queue1).
1.2.
OLT QOS ARCHITECURE
Interconnection between NT1 and LT is over point-to-point backplane links running at 10GE XAUI links in the system. These point-to-point interfaces are embedded in the backplane of the shelf, where an additional connection is foreseen from the LTs towards a second NT card, configured in loadsharing or redundant mode. Note that multiple network links can be used to connect the 7342 ISAM FTTU to the same peer system and therefore these links can be aggregated using 802.3ad. Up to two link aggregation groups are supported. Load balancing between different links within the Link Aggregation Group is supported.
Besides the backplane links, the NT has two 1-GE and two 10-GE ports to connect the system to the Ethernet MAN. Each of the upstream/downstream interfaces of NT – either network or backplane links – provide 8 QoS queues. Packet scheduling from these queues can be provisioned either in strict priority (SP) or weighted round robin (WRR) fashion. The weights of the WRR queues are also provisionable by the customer, based on how many contracts of each type (i.e. queues) have been signed and what is the operator’s policy for traffic class oversubscription. The NT software by default utilizes 4 out of 8 QoS queues. The first two queues are provisioned as SP and the last two queues are provisioned as WRR. Mapping of traffic class (802.1p) to QoS queue is customer configurable. In case of two SP queues on the same scheduler, the queue with the lowest number (Queue1 Queue8) will get the highest priority to be served. The LTs provide multiple gigabit interfaces via the backplane links to the redundant NT cards. No upstream scheduling functionality is foreseen at the GLT4-A board towards the backplane tracks as no bandwidth blocking is expected; one 10-G backplane link will be sufficient to deal with the four PON interfaces running at a maximal 1.25G upstream capacity. In the downstream direction towards the PON, the GLT4-A provides provisionable per-ONT or per-service and per-ONT rate limiting capabilities – as detailed in Error! Reference source not found..
OLT NT 1-GE queue1 … queue8
LT
SP/ WRR
LT 1
…
1-GE
LANX
queue1 … queue8
SP/ WRR
Ba ckpla ne XAUI link
10-GE
…
Forwarding and Downstream Rate Limiting @Traffic management module
Figure 4: OLT downstream QoS architecture Note that a generic packet switch buffering strategy is applicable to the NT and LT switching architecture. The switch employs a common buffer pool, which consists of a 4 Mega-Byte packet buffer for the NT card. This common buffer pool is integrated on chip and shared by all of the ports. The switch operates in a dynamic buffer management mode. The dynamic mode allows for static provisioning of a guaranteed portion of memory dedicated to each CoS per port, with the remaining memory in the buffer pool shared across the ports.
