I am aware that the Time Capsule is the only Apple-sanctioned solution for network backups. I am also aware that Time Capsules have a reputation for being unreliable (power supply failures mostly), and it's not clear what Apple has done to rectify this. Moreover, single mechanical hard drives are themselves ticking time-bombs, which is why enterprises prefer RAID solutions.
There are some NAS solutions, like QNAP, that claim to support Time Machine. I can say from personal experience that Netatalk isn't exactly the most robust software, so I'm not sure how I feel about trusting my backups to it. But this is one good and relatively inexpensive way to do TM backups to RAID.
While trying to come to a final decision, I've been using USB drives attached to an AEBS. One thing that's irritated me is that backing up any significant amount of data to this would crush my new 2011 17" MacBook Pro with 8GB of RAM. In this configuration, TM seems to be a massive memory hog. One hypothesis that occurred to me was that since TM is backing up to a sparsebundle, which is an encapsulated filesystem, uncommitted writes will be buffered, which will compete for RAM against applications. The sparsebundle is actually a set of files that are stored on the network volume, and that too will be buffered. Thus, every write will be buffered twice (at least). On top of that, since the host is much faster than the network, the channel to the disk will fall behind, consuming more and more memory as more unwritten data is buffered. Just a guess.
It never occurred to me that a real Time Capsule would be any different. But I was talking on IRC with a guy who claims to be friends with several Apple engineers. This anonymous person, whose word I take with a grain of salt, claims that the way backups are done to a TC is completely different to how they're done to a USB drive on an AEBS. He claims that a great deal of the overhead associated with managing the backup sparsebundle is offloaded to the Time Capsule. As a result, backups to a TC will impact the rest of the system much less. In researching this, I have come to find out that although you can export a TC backup bundle to an external drive, you cannot import one, partly because the on-disk formats are different. That Apple might use a completely different protocol to back up to a TC seems plausible and might explain some of the reported reliability differences between AEBS and TC setups. I'm assuming that AEBS and a NAS solution would really use the same protocol (AFP). Can anyone back up any of these claims?
Can anyone comment on the relative reliability of NAS, AEBS, and TC solutions in terms of data integrity?
What about the reliability of newer TC units in terms of hardware failures?
And how about performance?
If I've used Parallels, and it's taken a snapshot, many gigabytes will get backed up. When you encounter such a situation, what impact on system performance is there, and how long does it take to back up?
Thanks for the help!
There are some NAS solutions, like QNAP, that claim to support Time Machine. I can say from personal experience that Netatalk isn't exactly the most robust software, so I'm not sure how I feel about trusting my backups to it. But this is one good and relatively inexpensive way to do TM backups to RAID.
While trying to come to a final decision, I've been using USB drives attached to an AEBS. One thing that's irritated me is that backing up any significant amount of data to this would crush my new 2011 17" MacBook Pro with 8GB of RAM. In this configuration, TM seems to be a massive memory hog. One hypothesis that occurred to me was that since TM is backing up to a sparsebundle, which is an encapsulated filesystem, uncommitted writes will be buffered, which will compete for RAM against applications. The sparsebundle is actually a set of files that are stored on the network volume, and that too will be buffered. Thus, every write will be buffered twice (at least). On top of that, since the host is much faster than the network, the channel to the disk will fall behind, consuming more and more memory as more unwritten data is buffered. Just a guess.
It never occurred to me that a real Time Capsule would be any different. But I was talking on IRC with a guy who claims to be friends with several Apple engineers. This anonymous person, whose word I take with a grain of salt, claims that the way backups are done to a TC is completely different to how they're done to a USB drive on an AEBS. He claims that a great deal of the overhead associated with managing the backup sparsebundle is offloaded to the Time Capsule. As a result, backups to a TC will impact the rest of the system much less. In researching this, I have come to find out that although you can export a TC backup bundle to an external drive, you cannot import one, partly because the on-disk formats are different. That Apple might use a completely different protocol to back up to a TC seems plausible and might explain some of the reported reliability differences between AEBS and TC setups. I'm assuming that AEBS and a NAS solution would really use the same protocol (AFP). Can anyone back up any of these claims?
Can anyone comment on the relative reliability of NAS, AEBS, and TC solutions in terms of data integrity?
What about the reliability of newer TC units in terms of hardware failures?
And how about performance?
If I've used Parallels, and it's taken a snapshot, many gigabytes will get backed up. When you encounter such a situation, what impact on system performance is there, and how long does it take to back up?
Thanks for the help!
