I85PREP
⌘ K
Extreme Certified Professional - Extreme Switching study materials
Extreme Certified Professional - Extreme Switching study resources

Extreme Certified Professional - Extreme Switching: Link Aggregation and Load Sharing

Extreme Certified Professional - Extreme Switching · Work through your exam, one topic at a time.

All guide sections

Link Aggregation and Load Sharing

Domain 4 - Link Aggregation and Load Sharing. The exam checks that you can build static LAGs and LACP bundles on ExtremeXOS, know the rules a member port must satisfy, and predict which link a given flow will use.

Building a LAG

  • Static: enable sharing 1:1 grouping 1:1-1:4 makes port 1:1 the master and 1:2–1:4 the members.
  • LACP: enable sharing 1:1 grouping 1:1-1:4 lacp; at least one side must be active, set with configure sharing 1:1 lacp activity-mode active.
  • Configuration applied to the master (VLAN membership, STP, QoS, ACLs) is inherited by members. Configure the master, never an individual member.
  • Requirements: same speed and duplex, same media type, same VLAN membership, compatible LACP mode. show sharing and show lacp confirm state; members in collecting/distributing are usable.

Load-sharing algorithm

  • enable sharing 1:1 grouping 1:1-1:4 algorithm address-based L2 hashes MAC addresses; L3 hashes on source/destination IP; L4 adds TCP/UDP port numbers.
  • The hash is per flow, not per packet, so ordering is preserved. L4 spreads traffic better when many sessions share one IP pair; L2 still applies to non-IP traffic.

Worked example

A four-port LAG (1:1–1:4) connects to a server with a four-port team. show sharing shows all four ports active and show lacp shows all four distributing. show ports 1:1 statistics shows traffic, and 1:2–1:4 show traffic too but with different totals. A single backup job to one storage host saturates roughly one link, not 4 Gbps, because the flow's hash lands on one member; a second simultaneous flow to a different host spreads the load. Reasoning: the hash works on the flow's addresses (plus ports with L4), so parallelism comes from the number of distinct flows, not from combining their bandwidth.

Traps

  • Assuming a 4×10G LAG gives one flow 40 Gbps - aggregate capacity is 40 Gbps, but one flow stays at 10 Gbps.
  • Adding a port of a different speed or media type; that member silently fails to aggregate.
  • Configuring VLANs or STP on an individual member - the change is rejected or folded into the master, so the two sides disagree.
  • Passive LACP on both ends, so the bundle never forms.
  • Forgetting that STP sees the whole LAG as one logical port and never blocks individual members.