Why Time Machine is broken

Mac OS X's Time Machine backup is the easiest and cutest backup in the industry. But for many its performance stinks, forcing people to turn it off. Why?

A great idea Few people back up their data because it's work. You have to choose what software and hardware to use, buy it, configure it and then hope it works when you need it.

Time Machine solved that. Part of Mac OS X since Leopard it automagically backs up the stuff you care about.

It's easy: plug in a 2nd disk - USB or FireWire - to your Mac and it asks if you want to use that disk for Time Machine. Click yes and you're good to go.

TM backs up hourly. It keeps hourly backups for 24 hours, daily backups for a month, and weekly backups for all previous months thereafter. But it isn't an archive because it kicks old stuff out when the disk fills up.

To restore a file click on the menu bar TM icon, choose "Enter Time Machine" and you get the nifty interface:

(image courtesy Apple Inc.)

Civilians are always blown away by the Time Machine interface. It looks great and it works intuitively.

Problems People I support here in "the real America" started complaining about Time Machine's impact on performance. Spinning beach balls, long pauses, even system crashes.

I noticed it too and using TimeMachineEditor I cut the hourly schedule to once a day. It helped but finally I turned it off. I run daily non-TM backups and didn't need the grief.

But it made me wonder what was going on under the hood. Several factors contribute to Time Machine's poor performance.

• If a single block changes in a large file, the whole file is recopied. Expensive if you routinely work with large files.
• If you keep thousands of e-mails in your inbox - and I do - each Time Machine backup has to create thousands of hard links to each email.
• At the same time it has to break thousands of hard links up because the Time Machine starts deleting hourly backups after 24 hours.
• Hard links aren't costly in most Unix systems, but they are in Mac OS - each requires a couple of disk I/Os which kills performance.
• While new hard links are created, the system locks the file system B-tree, locking up the system.

The Time Machine engineers tried to make this overhead inconspicuous. If you're a light user you won't see much impact.

But if you receive and keep a lot of e-mail, download and keep a lot of content or perform large-file I/O intensive work - video, Photoshop, music - you'll find that Time Machine has a noticeable and perhaps unacceptable impact on system performance.

The Storage Bits take Casual Mac users are fine with Time Machine. But with Macs owning over 70% of the high-end PC market it is power users that Apple should worry about. They need Time Machine protection but they also need performance from their premium priced machines.

Time Machine's public face is wonderful. The intentions behind it are excellent. But the underlying infrastructure of hard links and complete file copies is incompatible with the high-end PC market.

Modern file systems use low-overhead copy-on-write snapshots, checksums and byte level change records to minimize storage I/O. TM's architecture, on the other hand, maximizes expensive disk I/O which kills machine throughput.

The obvious solution is for Apple to go back to their original plan to use ZFS as the new Mac OS filesystem. NetApp, whose patent suit against Sun scared Apple off, has settled with Oracle and would, no doubt, do likewise with Apple.

Bottom line: HFS+ isn't up to the job of meeting 21st-century file system needs. Apple needs a new foundation for the Mac OS and they need it now.

Courteous comments welcome, of course. Apple has a nifty intro to Tima Machine here.