AS7018 upstream loss via 32.130.88.22 (SF Bay Area) — per-flow, unidirectional

[mikecika--- via NANOG <nanog () lists nanog org>]

Seeing sustained upstream packet loss on AT&T Fiber (AS7018) in the SF Bay Area that looks path-dependent rather than 
congestion or CPE related. Posting in case anyone at AT&T can take a look, or has seen the same.

Symptoms:

Upstream collapses to ~2–5 Mbps on a 1 Gbps symmetric circuit
Downstream unaffected — 219 measured transfers, 0 failures
Host TCP counters show 5–14% retransmission on affected flows
Per-flow: back-to-back transfers to the same destination, differing only in source port, alternate between full rate 
and collapsed. Once a flow establishes it stays in whichever state it started in.

Affected destinations traverse 32.130.88.22 → 32.130.20.157. Destinations routing otherwise are unaffected (12 tested, 
0 failures).

Ruled out: CPE (3 gateways, 2 models), plan tier (5G → 1G), optics (no DDM alarms, PON O5), MTU (0% ICMP loss at all 
sizes), Wi-Fi (wired throughout).

~650 measurements over 7 days. Traceroutes and per-destination breakdown available on request.

Anyone else seeing this on AS7018 in the Bay Area?
_______________________________________________
NANOG mailing list 
https://lists.nanog.org/archives/list/nanog () lists nanog org/message/ESBPA7A4QTSLZZ6VQFYEPN25MQLKSV5L/