hoakley March 15, 2020 Macs, Technology

Last Week on My Mac: Backing up the future

It’s easy to forget – now that Time Machine is so popular among Mac users – that it was at least partly, if not primarily, intended to support Apple’s Time Capsule products. It’s nearly two years since Apple discontinued them, during which macOS has switched to booting from a new file system. Instead of quietly dropping Time Machine from macOS, Apple has instead re-engineered it to take best advantage of features in APFS such as snapshots.

That same period has seen great growth in cloud backup, and despite support for iOS devices backing up to iCloud, Apple hasn’t yet offered any similar macOS service to recoup the cost of its continuing investment in Time Machine.

More surprisingly, third-party backup software and services haven’t been put out of business by Apple. If anything, there’s an even wider choice today than there was back in 2007 when Time Machine first appeared in Mac OS X 10.5 Leopard. What’s worse for Apple, because of the requirements (self-)imposed by its App Store, all (serious) backup utilities are sold independently. Not only is Apple no longer generating any revenue from selling hardware which relies on Time Machine, but it’s not getting any service revenue from selling on third-party products either.

Some of Apple’s recent decisions in Time Machine development have brought big cost savings: when APFS snapshots were announced at WWDC, Apple revealed that this enabled it to strip some 10,000 lines of code which had previously been required to support ‘mobile Time Machine’, a marginal feature for mobile Macs which has been transformed into one of Time Machine’s strengths. But that still doesn’t generate revenue for Apple.

Perhaps the biggest obstacle to Apple generating service revenue from backups is the continuing dependence of Time Machine on directory hard links, apparently the sole reason that it still can’t back up to an APFS volume. This too is entirely self-imposed: Apple specified and implemented APFS for its own needs, but decided not to include this one feature which is still required by Time Machine’s backups.

This implies that Apple has already chosen a solution which will allow backups to be hosted locally on APFS, or in iCloud, encouraging many more Mac users to pay for 2 TB subscriptions, with the offer of greater in the future. But there’s a lot more to this than merely enabling sparse bundle backups in iCloud. To make this a good choice, it needs to be integrated with Recovery Mode so that full system recovery can be performed using remote backups.

If Apple were to integrate iCloud backup into what I have termed Time Machine 2, it would also be unlikely to steal many of those businesses which are already using cloud backup services. As we’ve been learning from your valuable comments here (thank you to all who’ve responded), those are generally carefully integrated à la carte solutions and often include servers and non-macOS systems.

For integrated iCloud backup to sell in volume to ordinary Mac users, it would need to be as simple as Time Machine is currently, with a fixed schedule and only basic options as to what is included, which would almost certainly not include anything now locked away on the Catalina System volume.

iCloud Time Machine also overcomes a major drawback with local backup to APFS volumes. Those of us who are fortunate enough to be able to mount more than 10 TB of SSD storage might whoop with joy, but for the average Mac user with a Time Capsule, external hard disk(s) or NAS, seeing any benefit requires substantial investment in replacement storage. As I pointed out previously, “who is really ready to back up to APFS just now?”

Many more Mac users could already take advantage of backing up to iCloud rather than to SSDs using APFS. Instead of having to replace their storage, as the size of their backups grew, they’d simply increase the size of their iCloud allocation, and the amount they paid Apple. Augmented subscriptions would bring Apple substantially more service revenue: ten million Mac users at an average of an additional $5 per month is an annual revenue of $0.6 billion. That would surely justify the engineering cost of Time Machine 2.

11Comments

Add yours

1

xz4gb8 on March 15, 2020 at 11:42 am

Perhaps someone at Apple has realized that Time Machine is an indirect revenue source, also called a sales enabler. My client base is consistently “not young”. The first older client that I helped into computing was only seventy years old and lived 100 miles away. The lack of worry about data loss provided by Time Machine allowed me to support an absolute “newbie”. Otherwise, I would have had to treat her like a Windows user and maintain my distance.

LikeLiked by 2 people
- 2
  
  hoakley on March 15, 2020 at 7:30 pm
  
  Thank you.
  I think, like any successful business, Apple looks hard at what investment in software engineering brings in terms of increased revenue. What we’ve seen recently in Time Machine has been good (apart from the bugs!), but has actually saved Apple maintenance costs in Mobile Time Machine. To progress to TM 2, with support for APFS hosting backups, is a much more substantial investment which isn’t actually needed: there’s no reason why TM couldn’t continue to back up to HFS+ for many years yet.
  It will be interesting to see what happens over the next couple of years.
  Howard.
  
  LikeLike
  - 3
    
    xz4gb8 on May 18, 2020 at 1:06 am
    
    I neglected to mention one important measure of the success of Time Machine — I have many older clients. In one case, I helped a 70-year old woman get her first MacBook. Her home is 100 miles away. I have visited her only twice in the seven years she has used the MacBook. I could never in clear conscience or in rational business sense advised her to get a Windows machine and I have full confidence that Time Machine allows me to increase Apple hardware sales because I can say, “What, me worry?”
    
    LikeLiked by 1 person
4

EcleX on March 15, 2020 at 10:26 pm

The reason is that APFS is not finished yet. Hopefully, we will see Time Machine 2 supporting APFS as target backup disks at WWDC by June 2020 and then released by September with macOS 10.16.

LikeLiked by 1 person
- 5
  
  hoakley on March 15, 2020 at 10:29 pm
  
  Thank you.
  Which features of APFS are being added to support TM then?
  Howard.
  
  LikeLike
  - 6
    
    Sebby on March 16, 2020 at 1:24 pm
    
    I think the most obvious “missing feature” is streaming deltas. If APFS had those, it would be possible to record and then apply filesystem changes directly instead of relying on whole file copies. Remember, when TM came out, it was rumoured that Sun’s ZFS would be in Mac OS X; even then it was an aberration and people were disappointed that the only optimisation was hard links to directories (which, incidentally, I am glad do not appear in APFS because they are very dangerous). Larry Ellison opened his big gob on the rumour and Steve Jobs then saw that it never happened, according to ZFS’s more ardent supporters. Certainly APFS has not been entirely useless because, as you say, the whole NFS-based mobile backup infrastructure is now gone and changes are read from snapshots. But until the target is APFS too, and assuming that Time Machine remains essentially the same product (i.e. transparent user access to files with no backup-level compression, de-duplication, direct non-disk-based or remote filesystem-protocol support, etc) I do not honestly see what else Apple can really optimise. Certainly if they are going to go for all this cloud-based malarkey, on which more later, I think they will need an approach far superior to simply trying to sync a sparse image of an on-disk filesystem to the cloud–they will need to be more like Arq, restic, or borg than simple filesystem copying to economise on the cloud and work with the kinds of small-disk-sized Macs we are now seeing, more than ever. In short I’m not sure Time Machine really has much of a future outside the designs of technology people and small businesses, which isn’t a big segment in a small market.
    
    I use restic myself, on my NAS, to back up to cloud storage; I use rsnapshot, which is basically Time Machine for UNIX-likes written in perl, to provide the filesystem transparency I need the rest of the time for local backups to the same file system on the NAS, which is itself backed by Linux RAID 6. I use Time Machine is a final safety net, in case I need to do a full recovery. Funny you mention Time Capsule; I just set up my old one for use again to take the load off my NAS because during a Sunday morning weekly RAID scrub, my Mac took several hours just to do a single backup and could not shut down during that time, and all because of slow response from the NAS during reads of bad disk blocks. Disastrous but that’s what happens with this kind of fragile abstraction. I’m sorry Apple got out of the router business–it seems clear to me now that this was a bad move because Time Capsule really did provide the goods, both Wi-Fi and (albeit, somewhat slow) backup disk. Clearly that would be the precondition for a revised Time Machine as well.
    
    Finally I’d like to take on iCloud for backup. The way iOS does it is obviously wrong: it only backs up user data, and Apple doesn’t consider it user data if you sync it from anywhere else, including your Mac. As a vehicle for lock-in to Apple’s media stores and iCloud services it cannot be beaten … but as a backup strategy, I can’t possibly endorse it. Stick to using iTunes, IMO. That way, you can also encrypt your backups–still no option for that with iCloud. Anyway, it would be unlikely to work on the Mac, because Apple doesn’t have any control over how and where user data is stored; their developer guidelines mean this is under complete control on iOS. You also need Keychain backup switched on if you are going to benefit from cross-device restores that are even halfway decent (because of the way keychains are stored and reliant on device keys, they cannot be restored directly to another device), and ten to one you’ll find that some app or setting still isn’t right. Clones, or complete restores, just don’t exist on the platform–something I’d infinitely prefer, even if I could safely skip an entire system OS partition, I’d only consider a cloud option if I could restore completely from it, on either platform.
    
    LikeLiked by 1 person
    - 7
      
      hoakley on March 16, 2020 at 4:32 pm
      
      Thanks.
      I don’t think streaming deltas are required for TM 2 to support backups on APFS, are they? I agree they’d be a valuable feature, but there are plenty of other ways around the issue. Snapshot deltas are already supported, as far as I am aware.
      Howard.
      
      LikeLike
8

J. G. Vía on March 16, 2020 at 7:51 pm

Reading your posts I can reconstruct your adventurous travel till finding your backup plan. And I also imagine Time Machine working on iCloud, as a service with a cost in a near future. Apple seems to keep interested in Time Machine, but seems they’re paying little attention to wired or private connection details in the software. Of course, they’ve forgotten about making dedicated hardware (a pity: my Time Capsule is an admirable always-works device). No more granular and “pro” control in Time Machine, neither.
BUT it’s alive!
After watching you getting into the guts of backing up, I come to agree about a cloud solution for Time Machine. And a personal mix of procedures depending on needing for us users.
(I keep using modified Time Machine in Mojave for one backup a day. And a weekly bootable copy on hard drive. I have two of them, and I take newest to my mother-in-law’s; I think if a lightning strikes her house, it’s unlikely it’d also happen to me).

LikeLiked by 1 person
- 9
  
  hoakley on March 16, 2020 at 11:33 pm
  
  Thank you.
  I used to keep my off-site backups on DVDs in my parents’ house around 12 miles away. When my father died late last year, and my son was clearing the house, it was quite spooky to get all those old DVDs back.
  Howard.
  
  LikeLiked by 1 person
10

Sebby on March 17, 2020 at 11:32 am

Hi Howard.

AFAICT, at the moment Catalina using ASR only supports direct cloning of deltas between two live APFS filesystems; you can’t record the deltas for storage into a file or transmission across a network for reconstruction on some server machine either on a clone of the filesystem or at a later time when restoring. To my mind, this ultimately begs the question: why APFS then? If one of the most important new features of Time Machine (efficient incremental updates) can’t be done without either a proprietary server or a directly-connected APFS disk (which, as you say, is going to be expensive while SSDs are a better fit for that filesystem), and if the gains are only really marginal anyway, then why not just have a much more protocol-agnostic and filesystem-agnostic chunk-based backup technology like that in Arq or restic that Apple can optimise for maximum robustness, search and browsability? Those are going to be better for the cloud, and are much more likely to be the basis of an enhanced iOS backup system if it ever emerges. I think the use of APFS is a very marginal gain indeed. I can’t see a future use for the fragile sparse bundles if backups keep getting bigger, and as you say Apple will have to enhance the backup of the snapshots with much more than just the deltas themselves. I think it’s going to be a rethink, and anyone who wants APFS on direct disks just goes out and buys CCC which now has the feature (implemented using ASR).

LikeLiked by 1 person
- 11
  
  hoakley on March 17, 2020 at 5:05 pm
  
  Thank you. I think that’s correct, but I don’t think that’s a limitation in APFS, rather what ASR can do with it at present. If Apple is going to support any backup destination apart from HFS+ and sparse bundles in the future, that’s exactly the engineering work I’m referring to here. But I don’t think that requires any significant changes to APFS itself.
  Howard.
  
  LikeLike

·Comments are closed.

Share this:

Like this:

Related