Two MacBooks, 10.11.4, one quick TimeMachine backup and one very slow

Question

Level 1

9 points

Two MacBooks, 10.11.4, one quick TimeMachine backup and one very slow

In our home office, connected to a Gigabit ethernet network on the same switch, we have two MacBooks. One is a 2011 MBP with a 1 TB SSD (third party), and one is a 2015 MBA with a 512 GB SSD (Apple). A Linux server on the same network runs Netatalk 3 providing the network backup drives, one for each laptop. The two TImeMachine network drives are on the same physical server drive. Both laptops are running 10.11.4 and are configured pretty darn close to identically. Neither laptop has any trouble finding the network backup drives, and they start running backups without problem.

The old MBP has been a workhorse and running flawlessly for years and years, including fast and regular TimeMachine backups. Single-file restores are done with some regularity, and they have always worked (I think only once did we use a full system restore from scratch, and it worked).

The new MBA, on the other hand, has had endless issues starting and completing backups. Last night, for example, I decided to simply reformat the drive on the server in case there was a hardware error affecting the MBA (I ran Linux fsck to check the filesystem before wiping, and it found no errors), and started full backups on both laptops all over again. The MBP has about 600 GB for the first backup, and it completed the backup in about 5 hours.

The MBA, with about 350 GB to back up, was still running this morning, and had completed less than half its backup. From the console log, it showed it was backing up about 5 GB/hour. Not only that, but incremental backups have been taking many hours (more than overnight), making it impossible to ever complete them because it has to get packed up and taken to work at some point in the morning. I.e. backups are just plain broken on this MBA.

The MBA connects to the network with a third party (Anker) USB3 hub that has a Gigabit Ethernet port on it. I downloaded iperf for the Mac and the Linux server and found that throughput is right about 900 Mbps, so I expect full bandwidth is achieved through that connection.

Any ideas what might be going on here? Diagnostics between the two to see what might be such a radical difference? Have tried changing network taps, in case of wiring/switch issues, no change. Rebooting, no change.

We have two other Macs also running backups this way, under 10.11.4, and they have also been problem-free. So it's just this one MBA, the newest one of the bunch, having problems.

MacBook Air, OS X El Capitan (10.11.4)

Posted on Apr 29, 2016 12:13 PM

Reply

Answer 1

Linc Davis

Level 10

209,547 points

Apr 29, 2016 12:37 PM in response to mklein9

A third-party file server is unsuitable for use with Time Machine, especially if it's your only backup. I know this isn't what you want to hear. I know that Time Machine accepts the device as a backup destination. I know that the developer says the device will work with Time Machine, and I also know that it usually seems to work. Except when you try to restore, and find that you can't.

Apple has published a specification for network devices that work with Time Machine. No third-party vendor, as far as I know, meets that specification. They all use the incomplete, obsolete Netatalk implementation of Apple Filing Protocol.

Apple does not endorse any third-party network device for use with Time Machine. See this support article.

If you want network backup, use as the destination either an Apple Time Capsule or an external storage device connected to another Mac or to an 802.11ac AirPort base station. Only the 802.11ac base stations support Time Machine, not any older model.

Otherwise, don't use Time Machine at all. There are other ways to back up, though none of them is anywhere near as efficient or as well integrated with OS X. I don't have a specific recommendation.

If you're determined to keep using the device with Time Machine, your only recourse for any problems that result is to the developer (which will blame Apple, or you, or anyone but itself.)

Reply

Answer 2

mklein9 Author

Level 1

9 points

Apr 29, 2016 3:00 PM in response to Linc Davis

Linc, I greatly appreciate your response. I have previously read your response to others' questions but I decided to ask anyway. Note that the NetAFP site you link to talks about Netatalk 2.2 and I did mention I am running Netatalk 3 (3.1.8 specifically, which is the most recent release). I am not using a third party NAS box but a fully updated Debian-based server running the latest Netatalk.

In any case, for 3 of our 4 Macs we have years and years of excellent experience with Netatalk as a TimeMachine server. It is only this one MBA, running the same Mac OS X as the others, that has been different.

Reply

Answer 3

Linc Davis

Level 10

209,547 points

Apr 29, 2016 3:06 PM in response to mklein9

I have no experience using Netatalk with Time Machine. My opinion is based on many reports of data-loss catastrophes that happened to others who did what you're doing—months or years of apparently trouble-free backups, then when something had to be restored, total failure with no warning and no way to recover. If you want to follow in their footsteps, then you are, as I wrote above, on your own.

Reply

Answer 4

mklein9 Author

Level 1

9 points

Apr 29, 2016 3:14 PM in response to Linc Davis

Linc, I know I'm on my own, never thought otherwise. I've been managing our home network for about 20 years and wouldn't have it any other way. Zero data loss, fast unlimited email, reliable file systems 24/7, except for TimeMachine backups on one specific MBA. Just looking for diagnostic hints, if any, on this topic.

Reply

Answer 5

mklein9 Author

Level 1

9 points

May 4, 2016 11:14 AM in response to mklein9

More info. Next step: try the same backups to an external USB3 drive, freshly erased and initialized. Same exact symptoms. The initial scanning and first number of GB run fast, as expected, then it hits a patch of molasses. The backup rates are essentially identical whether backing up to the network Time Machine server or an external drive: 5-6 GB/hour. So it is unlikely to be a network or AFP problem, but a USB3 problem. Again, 3 of 4 other Macs on the same network run network Time Machine backups without any issue, and all are running 10.11.4.

I ran across this posting on StackExchange noting that high I/O consumption over an extended period of time may cause backupd to be throttled: http://apple.stackexchange.com/questions/212537/time-machine-ridiculously-slow-a fter-el-capitan-upgrade. I tried the suggested fix of disabling throttling, and it did not help either with network or external HD backup. All the other Macs on the network have the same throttling setting enabled and they do not have issues.

I've also tried forcing a file system check with "fsck -fy" in single user mode, and it said the Macintosh HD volume appears to be OK.

In both network and external HD cases, I see a lot of I/O, but very little data is appearing on the backup device. While 5-6 GB/hour are being backed up, there is about 20-40 MB/s of I/O to the backup drive, but nothing seems to show up there, or very very slowly. Here is a snippet of iostat output; disk2 is the external Time Machine USB3 drive, while the Time Machine window is frozen at backing up 2.8 MB out of 100+ GB:

root# iostat -d 10 disk0 disk2 KB/t tps MB/s KB/t tps MB/s 68.50 135 9.01 85.83 173 14.46 114.22 1 0.10 0.00 0 0.00 117.21 181 20.72 123.78 322 38.94 54.94 333 17.89 127.21 103 12.74 105.52 247 25.50 126.61 33 4.10 116.76 450 51.33 126.98 135 16.79 14.91 6 0.08 116.07 91 10.28 6.15 1 0.01 125.78 78 9.63 11.83 8 0.10 126.53 528 65.25 10.40 0 0.01 128.00 530 66.19 6.87 5 0.03 98.86 223 21.49 11.83 8 0.09 88.64 180 15.57 0.00 0 0.00 98.19 206 19.71 6.13 1 0.01 103.75 241 24.42 8.15 15 0.12 101.66 215 21.30 50.83 340 16.88 120.21 167 19.64 120.78 172 20.26 126.07 18 2.19 121.49 392 46.56 122.55 14 1.67 100.59 79 7.80 124.83 339 41.38 124.37 287 34.90 127.17 56 6.99 disk0 disk2 KB/t tps MB/s KB/t tps MB/s 11.38 10 0.11 122.17 202 24.15 21.60 0 0.01 125.56 92 11.24 6.52 3 0.02 127.56 439 54.65 11.24 10 0.11 127.17 540 67.02 8.92 11 0.09 84.82 172 14.29

Reply

Answer 6

mklein9 Author

Level 1

9 points

May 9, 2016 9:36 AM in response to mklein9

To anyone following this discussion but not the newer, related one I started with new info here: https://discussions.apple.com/thread/7554773, here is what appears to be the resolution. Bottom line, a new MacBook Air, and a possible hint about turning off Spotlight indexing while at least the initial TM backup is going on.

Here is the resolution copy-pasted from the thread linked to above.

==============================================================================

I think I have solved this, but see the detail to determine whether it might work for you.

First of all, given the detailed experiments and data I had, I went to the local Apple Store Genius Bar and they rather quickly diagnosed it as a likely hardware problem related to the SSD. In the end, I walked out of the store with a new MacBook Air.

Came home, restored the new MBA from the one and only TM backup I'd ever gotten to finish in the last two months (on the external USB3 drive), which worked great. First order of business was to start a network TM backup. Initial 50 GB again went fast (about 1 GB/minute) and then started slowing down, approaching 5 GB/hour.

One of the things the Genius pointed out was if high CPU usage of mds_stores was identified, it could be a problem with the Spotlight indexing service. On a hunch, while TM backup was slowing down, I disabled Spotlight indexing on all volumes with the command:

sudo mdutil -a -i off

and pretty much immediately the backup rate came right back up to, and exceeded, the previous 1 GB/min rate. At this rate, after 4 hours the full backup, 280 GB, was done, and the first incremental after that took 2 minutes.

Afterwards I reenabled Spotlight, but only on the local drive, with:

sudo mdutil -i on /

It remains to be seen if this is a permanent fix or not, but it certainly brought the full initial TM backup time right into the range it should be, limited largely by server drive write bandwidth. In any case, Spotlight indexing network shares, at least in our case, should be considered carefully as it may be extremely flaky. See the post https://discussions.apple.com/thread/6499628?answerId=26520737022#26520737022.

Reply