Best Disk Format for Time Machine on Monterey

Yes, SSDs have massive performance benefits over HDDs when APFS is used, but Time Machine is a special case scenario. IMO, the performance difference really doesn’t matter for most people because:

  • TM backups run once per hour, and in the background. As long as an hourly backup completes before the next one starts, who cares if it finished in 5 or 10 or 15 minutes?

  • APFS’s ability to create massive fragmentation is somewhat mitigated by the fact that it isn’t randomly modifying files. It creates a snapshot and then writes the files that changed. So changes tend to be localized near each other. Even though periodic snapshot deletion (hourlies after 24 hours and dailies after a week) does create fragmentation, it’s not going to be nearly as much as you might see on an HDD used as a boot device or for document storage.

  • The price difference is likely to be more important to those of us on a budget (almost all of us), especially when the use-case is backup and not for working files.

    For example, a 2TB SSD costs $130-280 (from MicroCenter). Looking for an external HDD for under $150, I found a 6 TB drive). Looking for an external HDD for under $280, I found a 16 TB drive).

    The ability to get 3-8x the capacity for the same price, especially for an application where performance is not critical, is not something you should ignore when making purchasing decisions. (Of course, you should still research each products’ reliability before making a purchase and not just look for the lowest price/TB ratio. This is true for both SSDs and HDDs.)

1 Like

As I understand Howard’s reply, the cumulative fragmentation caused by file deletions within snapshots becomes significant after an extended period of time.

As much as I respect Howard’s insight, knowledge, and recommendations (I’ve cited/referred to him quite a few times here on TidBITS), my experience with using hard drives formatted with APFS is not consistent with what he’s saying currently. I’ve been using two 4TB hard drives formatted with APFS for TM backups, and two other 4TB hard drives for CCC backups, for nearly two years. The CCC backups are considerably faster than when the drives were formatted for HFS+, and the TM backups have not given any indication of any significant slowdown (and in fact, seem to be somewhat faster) in that period of time. None of the drives show any indication of excessive fragmentation (though perhaps they have more than when HFS+ formatted).

I also don’t agree with his stated expected lifetime of a hard drive as being 2-3 years, a timespan that in my experience, is ridiculously low: several of my drives are at least 10 years old yet are doing fine. I’ve had three drive failures in the last 20 years. I feel I have sufficient redundancy (two independent backup systems, not counting an online backup, using alternate backup drives) to account for the inevitability of future drive failures. In all, I have 12 hard drives in use across the four systems I’m responsible for. The drives used for file storage (11TB of media files) are formatted HFS+. The drives used for backup (TM and CCC) are formatted APFS.

I don’t think a 1TB drive, or even 2TB, is adequate for backing up nearly 1TB’s worth of files. 4TB SSDs are still obscenely high in price and likely to remain so, notwithstanding what is now many years of being on the market. So I’m supposed to replace (in my case) at least four 4TB hard drives with some clutch of 8 to 16 SSDs with their far inferior price/TB ratio when the hard drives are still performing well? I don’t think so.

All above assume “low speed” SSDs (SATA/USB 3 etc). Howard mentions his hopes for prices of Thunderbolt front end SSDs to come down. I have the same hopes, but the truth of it is, Thunderbolt SSDs are 8x or more the prices of their USB 3x counterparts, and given the fact that we’ve been waiting for Thunderbolt prices to come down since the intro of Thunderbolt 1, we’re going to be waiting for a long time to come. I fully agree with Shamino’s recommendations regarding price/storage ratio.

I’m still a huge fan of Howard’s and consider him at the top of the heap in Mac expertise. His contributions to the community are immense and ongoing. However, there is a small subset of items where my results seem to vary a bit from his…

2 Likes

Maybe so, but even he says that this ultimately boils down to performance. And for something like Time Machine, performance really isn’t critical, as long as each backup completes before the next one is scheduled to begin.

But I would suggest looking for a high performance HDD. A small 2.5" portable drive is not going to perform anywhere near as well as a 7200 RPM 3.5" drive designed for 24x7 server operation (e.g. a “NAS” or “Enterprise” drive).

You may have to build your own external drive in order to get this, because most sold-as-external drives have slower consumer-grade drives in them. But that shouldn’t be a problem - it is very very easy to buy a bare drive and a USB enclosure and mount one in the other.

I’ve experienced the same thing.

I assume this is because the act of creating and deleting snapshots is much much faster than creating and destroying hard-links to duplicate files/directories (for TM) or moving files to a SafetyNet folder (for CCC), and not because of the nature of APFS itself.

But for a user (vs. a kernel developer), the reason for the better performance doesn’t matter. The fact that the APFS backups perform faster is all that matters.

I don’t believe for an instant that this observation would hold true for a general purpose volume holding actively-used apps and data, but backup media has a use-pattern that is very different from general-purpose access.

It also helps to have backup drives that are substantially larger than your source drive(s), so thinning doesn’t happen for a long time. In my case, my internal SSD is 2TB, of which the Data volume (that gets backed up) is about 1TB. I back it up to a 4TB HDD which currently has backups going back to May 11 (7 months - when I wiped it to switch from HFS+ to APFS) and has 2.6 TB available.

Please tell me more. I have noted in other threads that Time Machine has become wonky after two recent upgrades to macOS, and I think (without certainty) that an attempt to backup to my Time Capsule might well take over an hour (at which time a new snapshot should presumably begin).

On the other hand, Time Machine backups to an external spinning HDD take under an hour and they still fail, so it is probably something unrelated to a new backup colliding with a backup in progress.

One other thought. I believe, back when my TM backups worked, that the time shown for the most recent backup was the completion time of the backup. If this is the case, then it seems like the “next one” would be scheduled for an hour after the last one completes, and there could never be a collision.

I’m not sure what else I can add, but some possibly useful points:

  • A Time Capsule is not like a directly-connected storage device. It is network storage. As such, TM will back up to a “sparse bundle” disk image. This image may be (I think) formatted as HFS+ or APFS. I don’t know which is used by modern macOS installations, but if it is treated as locally-attached storage, then it is determined by the version of macOS that created it. Catalina and older creates HFS+ TM volumes, while Big Sur and later creates APFS TM volumes. If you created your TM volume as HFS+ (with an older macOS), it is not upgraded to APFS, but remains operating as an HFS+ TM volume until you decide to delete and re-create it.

  • I’m not surprised that a TM backup to a Time Capsule takes longer than to a local HDD. USB3 has a higher theoretical top speed (5 or 10 Gbps, vs. Ethernet’s 1Gbps), and network protocols like SMB are going to have more overhead than something like UAS, which most good quality USB3 enclosures support.

  • TM’s reliability seems to be different for different people. I never had a problem with it on my systems, although I did experience the bug (on Catalina, I believe), where it would stop making automatic hourly backups after a few days of uptime. I worked around this by manually telling it to run a backup whenever I noticed it hadn’t run for a few hours, which seemed to get it working properly for the next 2-3 days.

    My current Mac (running Big Sur) hasn’t given me any problems.

  • No, there can never be a collision. If a backup is running when the time comes for another, then the new one doesn’t happen (or gets delayed, I don’t know which). My point was simply that you don’t need backups to complete very quickly, as long as it’s fast enough that your hourly backups remain approximately an hour apart.

That’s what I would have assumed, but it seemed like you implied otherwise.

I understand (and agree) that performance is not a high priority, and that using a Time Capsule includes a lesson in patience.