Meta production RoCE fabric overprovisioning (under-subscription): rail-switch uplink capacity first doubled to run 1:2 under-subscribed against ECMP flow collisions, later trimmed to 1:1.125 — deliberately not 1:1, to keep buffer for up to two link failures
1:2 → 1:1.125observed
| Value kind | observed — Reported measurements, counts, and specifications keep the precision and scope stated by their source; an exact specification is not treated as a range. |
|---|---|
| As of | 2024-08 |
| Source | Gangidi et al., 'RDMA over Ethernet for Distributed AI Training at Meta Scale', ACM SIGCOMM 2024 (Meta) · §4.1.3 (1:2 under-subscription after doubling RTSW uplinks) and §6.2 (Stage 4 reduced to 1:1.125; not 1:1 to allow buffer for up to two link failures) |
| Review | checking…review by 2027-10-17 · fast cadence |
| Recorded changes | last 2026-10-17 · 3 revisions tracked |
| Claim id | meta-rtsw-under-subscription-ratio |
Where the guide uses it
← Full numbers register — every date-stamped figure in the guide, with revision history.