Back to matrix

G05-BA: No

Rationale

The selected Linux 6.12 native BATMAN_IV mesh lacks a credible global-state partition for the common ten-billion-device setting. Source Ethernet traffic registers/refreshes each attached client in the local translation table; the default inactivity timeout is 600 seconds. A smooth schedule of 1000 accesses per device per day, spaced 86.4 seconds apart with staggered phases, keeps clients active without burst pressure. Remote translation changes are synchronized through OGMs and table requests, independently of local destination demand. The 64-bit field-only lower bound from round431 is 54 bytes per represented remote client: a near-global replica costs approximately 540 GB before padding, originator lists and other state. This state is repeated at mesh forwarders, including access equipment, rather than partitioned by region or the 1000 destinations each client uses. Adding ordinary mesh forwarders does not shard the remote table. The local table-size guard rejects new clients whose table would exceed the supported response size; it is not a global aggregation mechanism. Together these are an architectural/resource obstruction for this selected native deployment on the common contemporary equipment envelope.

Assessment method

Architecture/state-lifetime trace and conservative memory accounting; no runtime benchmark

Source context

Pinned Linux 6.12 source and common population/workload; reviewed 2026-10-22

Outstanding qualification

Source-based plausibility verdict, not a measured global benchmark or a claim that batman-adv is unsuitable for local meshes. The evenly spaced schedule is an explicit ordinary-use realization of the shared daily workload, not an observed Internet trace or a new mandatory timing rule. The memory bound applies to concurrently represented remote attached clients; local clients are subtracted. It does not multiply identities owned by daily destinations. Moving devices into mesh-originator roles also requires per-originator reachability state (G01-BA), rather than eliminating global state. Bridge-loop avoidance suppresses duplicate backbone advertisements; it is not a selected inter-region namespace partition. Gateway mode is off and an IP hierarchy, NAT or another inter-mesh routing layer would change the selected stack. The 64-GB figure remains a sizing reference: the finding is widespread replicated state, not failure of a newly imposed simultaneous benchmark threshold. No allocation stress, packet loss, throughput or CPU measurement was run.

Public sources

Editorial evidence references

The following identifiers refer to research records not included in this public package; they are not public links.