Lag On Access - Alcatel-Lucent 7750 SR OS Interface Configuration Manual

•
•
•
For all cases that involve per-packet hashing, the NPA produces a 20-bit result based on hashing
the relevant packet data. This result is input to a modulo like calculation (divide by the number of
routes in the ECMP and use the remainder) to determine the ECMP index.
If the ECMP index results in the selection of a LAG as the next hop, then the hash result is hashed
again and the result of the second hash is input to the modulo like operation (divide by the number
of ports in the LAG and use the remainder) to determine the LAG port selection.

LAG on Access

Link Aggregation Groups (LAG) is supported on access ports. This is treated the same as LAG on
network ports which provides a standard method to aggregate Ethernet links. The difference lies in
how QoS is handled. If all members of the LAG are on the same IOM then there is no difference in
how HQoS is handled. For example, for routed packets or for VPLS known unicast, hashing is
performed to get an ECMP value and then rehashed to do the spraying for the access LAG ports.
SAP to SAP VLL, and VPLS flooded packets will only use one port of the LAG.
There are two user-selectable modes to address the need to manage an HQoS policy on a SAP (this
can include a link aggregate which spans cards).
1) Divide the SLA among the IOMs, based on their share of the LAG group. For example, a 100
Mb PIR with 2 links on IOM A and 3 links on IOM B, IOM A would get 40 Mb PIR and IOM B
would get a 60 MB PIR. The advantage of this method is that the overall SLA can be enforced.
The disadvantage is a single flow cannot exceed the IOM's share of the SLA. This is the default
method.
2) All ports get the full SLA. With the example above, each port would get a PIR of 100 Mb. The
advantage of this method is a single flow can consume the entire SLA. The disadvantage is that
the overall SLA can be exceeded if the flows span multiple ports.
7750 SR OS Interface Configuration Guide
→ VPLS multicast traffic transmitted on SAPs with IGMP snooping enabled is load-
balanced based on the internal multicast ID which is unique for every (s,g) record.
This way, multicast traffic pertaining to different streams is distributed across
different LAG member ports.
VLL traffic from a service access point is not sprayed on a per-packet basis, but as for
VPLS flooded traffic, the service ID is used to pick one of the ECMP/LAG paths. The
exception to this is when shared-queuing is configured on an Etherpipe SAP or Frame
Relay pipe SAP. In this case, traffic spraying is the same for VPLS known unicast traffic.
IP multicast is sprayed over LAG based on the unique multicast ID.
IP multicast Layer 4 UDP traffic will not be hashed.
Interface Configuration
Page 73