Hedronite · Cert Lesson · Cert-Prep / AWS SAP · Tue 2026-09-22

AWS SAP Direct Connect resiliency locations and LAG

Two ports in one building is not two buildings. A LAG adds bandwidth at one DX location; it does not invent a second site.

Lesson Class: Cert (AWS SAP-C02 · Direct Connect)
Exam Facet: Resiliency ladder + LAG same-location trap
Paired Ops: Python SQS DLQ and redrive policy census
Paired Dev: Python functools.singledispatch and register
Grounding: SAP direct-connect.md Resilience/HA + LAG
Ladder
Single → dual-port → dual-location → extreme.
LAG
Same DX location · Active/Active · minimumLinks.
Trap
LAG is bandwidth, not metro HA.
Second location before fatter LAG.

<!-- hal:authoritative:yaml -->

Two ports in one building is not two buildings. A LAG adds bandwidth at one DX location; it does not invent a second site.

§I — Frame

Last SAP fire on SoT taught Interface VPC endpoints and PrivateLink (09-07). That leftover stays on its shelf. Route53 failover (08-26) and Transit Gateway RAM (08-14) stay too.

Today the exam grain is hybrid physical path resilience. SAP bootcamp: order two DX ports, prefer two DX locations, put customer routers and on-premises routers in geographically separated buildings. Then contrast LAG: multiple physical connections acting as one, same speed, same DX location, Active/Active, minimumLinks for when the bundle stays up.

§II — Resilience ladder (bootcamp images 1 through 4)

PatternWhat you buyFailure it still eats
Single port, single locationOne LOA-CFA cross-connect pathDevice, cage, or building loss
Dual ports, one locationTwo cross-connects, two customer DX routers into two on-prem routersWhole DX location / metro hit
Dual locationsTwo DX locations + geographically separated routersRegional design mistakes on VIF/TGW side remain
ExtremeDual locations plus redundant on-prem buildings and routingStill needs correct VIF (Private/Public/Transit) and BGP hygiene

Bootcamp labels the progressive diagrams: none, resilient, improved, extreme. The exam trap is calling a single-location dual-port design "multi-site HA."

Physical facts that feed the design: DX is a port allocation at a DX Location (1/10/100 Gbps dedicated, or partner hosted speeds). Single-mode fibre. BGP + BGP MD5 on the customer router. Provisioning measured in weeks or months. Inbound data transfer free; outbound charged. LOA-CFA authorizes the colo staff to plant the cross-connect.

§III — LAG: bandwidth tool, not site insurance

Link Aggregation Group:

Exam contrast pair:

  1. Need more Gbps at one colo → LAG (or a faster port).
  2. Need survive colo or metro loss → second DX location (and usually second on-prem path), not a fatter LAG.

MACsec (hop-by-hop L2 encryption on the DX/LAG segment) is adjacent, not the resiliency answer. VPN over Public VIF is end-to-end encryption; MACsec is not a substitute for that story.

§IV — VIF adjacency (do not reteach PrivateLink)

Resiliency of the physical path is useless if the logical overlay is single-homed wrong:

Dedicated DX: up to 50 public/private VIFs and 1 Transit VIF (bootcamp). Hosted VIFs share into other accounts. Today only names these so the resiliency answer does not accidentally become a VIF type question.

§V — What not to do

Do not answer "add a LAG" when the stem says the DX location failed. Do not treat dual ports in one building as extreme resiliency. Do not confuse DX (private physical path, weeks to provision) with Site-to-Site VPN over the internet (hours, encrypted, variable latency). Do not re-teach Interface VPC endpoints / PrivateLink (09-07) as tonight primary. Do not open Route53 Resolver hybrid DNS as primary (valid future SAP fire; not this bundle).

§VI — Close instruction

Sketch four boxes: none / dual-port / dual-location / extreme. Write one sentence each on what dies. Then write the LAG one-liner: same location, Active/Active, minimumLinks, bandwidth not metro HA. Pair: Ops censuses SQS DLQs; Dev registers singledispatch formatters. Maghrib owns the quiz later.