Time Machine Wi-Fi backups fail on modern Macs

I had a 2012 Mac that I upgraded to an M5 Max. I dealt with Rosetta going from Power PC to CISC and my basic assessment was CISC->RISC is the same - Rosetta doesn't solve anything, it only delays the inevitable: your codebase needs to be recompiled or it will just die on the vine. After struggling through that whole migration, I had another with Time Machine.

I had a Time Capsule that I upgraded with a 10 TB drive years ago and recently decided to decommission it. Then I got the M5 Max and the new Apple Silicon does not play well with legacy apps, so I removed all of those. Then I found another problem - consistent failures with Time Capsule across many (days) trials over different hardware. I think the crux of this falls on Apple, and I asked Claude to clarify.

Me:

Agreed-it's a pretty *** interface but it has proven it's value through the years [referring to a plug-in USB3 to Ethernet adapter]. I am sort of confused by the failure on wifi, because in the past with AFP and a time capsule, everything went, just slowly, without ethernet. Now with Wifi 6 and 7, things seem broken. Not sure why things would degrade so recently compared to 15 year old tech.

Claude:

It's a great question, and the answer is genuinely counterintuitive: the newer stack is faster but far less tolerant, and Time Machine's network problem was never about bandwidth. Three things changed underneath you, none of them the radio.

1. The protocol: AFP → SMB. Your Time Capsule era ran AFP, Apple's own filesystem protocol — native, Mac-tuned, and forgiving. Apple deprecated AFP years ago; SMB is now the only supported network Time Machine transport. SMB is a Microsoft protocol Apple ported, and on a third-party NAS it's a whole different codebase (Synology's oplocks, SMB multichannel, signing) versus the old Apple-tuned-both-ends Time Capsule. You lost the home-field advantage.

2. The filesystem inside the bundle: HFS+ → APFS. This is the big one. The old sparsebundle held HFS+, which is loosely journaled and latency-forgiving — a stalled write just waited. Modern Time Machine uses an APFS sparsebundle, and APFS is copy-on-write and aggressively transactional: every commit needs the underlying storage to honor write barriers and flushes in order. Over a network, that means each transaction is gated on round-trips completing reliably. A jittery link doesn't make APFS slow — it makes it stall, because a barrier that doesn't return on time wedges the whole transaction. That's precisely why it hung at MountingDiskImage: mounting an APFS sparsebundle means reading and validating container checkpoint metadata over the wire before it'll mount, and if the link hiccups mid-validation, it just sits there. HFS+ never demanded that.

3. The radio got busier, not just faster. WiFi 6/7 gave you huge peak throughput, but TM cares about latency consistency, and modern Macs actively work against that:

• AWDL / peer-to-peer coexistence. Your Mac now runs AirDrop, Continuity, Handoff, Sidecar, etc. on a time-sliced peer-to-peer channel that shares the same radio as your infrastructure WiFi. That slicing causes periodic latency spikes — which is literally the en0 p2p "datapath timeout, packets stuck 12 minutes" we saw in your logs. 2010-era WiFi didn't do any of that; the radio just talked to the base station.
• WiFi is still half-duplex and contention-based — under a sustained bidirectional transactional load, latency goes bursty exactly when APFS-over-SMB can least afford it.

So the paradox resolves cleanly: old = slow but tolerant (AFP + HFS+ would grind through a flaky link because nothing was strict), new = fast but brittle (APFS + SMB + a contended radio, where the crash-consistency guarantees that make APFS safe are the same ones that make it stall on an unreliable transport). Your gigabit USB dongle wins not because it's faster — it's slower than WiFi 7's peak — but because it's steady: no jitter, no p2p time-slicing, no retransmits. Transactional filesystems love boring, consistent latency, and wired delivers exactly that.

It's a real regression in robustness, and you're not imagining it — Apple traded the forgiving old path for a stricter, snapshot-capable one and bet on the network being reliable. On WiFi, that bet doesn't hold for a big sustained seed; on wire, it's fine.

[Edited by Moderator]