Time Machine very slow

[Andy]

Hi
I've recently set up my DS218+ with DSM 7.01 to backup my three Macs via Time Machine with SMB. I'm pretty sure I've done it all correctly as it's all working. But it's incredibly slow. The initial back up was not slow, but subsequent backups are really slow- Over an hour for 600MB.
Im wondering about the 'Enable File Fast Clone' option in File Services/Advanced? Is it an option for Time Machine? Any other ideas why it's so slow? (not my connection by the way- I'm in my LAN on stable fast WiFi)
Thanks

 

[Seaprobe]

Fast clone won't make any difference as network backups are to a disk image file. The NAS has no idea what is inside that image.

Also, WiFi tends to be noticeably slower than ethernet for backup. The nature of TM backups is a very large number of small writes and the higher latency of each of those transactions adds up very quickly.

The biggest thing that helped for me was upgrading the RAM in my Synology. DSM quickly used that for cache, presumably for filesystem structure which is continuously updated, and made a dramatic difference. I found that the hourly TM backups were taking longer than an hour with stock RAM whereas and after the RAM upgrade were down to about ten minutes.

 

[Andy]

Great, I hadn't thought of RAM, I'll definitely do that. And thanks for the info on Fast Clone.

 

[Andy]

Actually..I'm looking at the Resource Monitor in the middle of another 2 hour epic backup of 173Mb and the RAM usage is only 37%? So it looks unlikely to be a RAM issue..unless I am not understanding.
(actually make that 4 hours for 173MB!)

So now it is stuck on 'Cleaning Up' for about 3 hours.

What is going on?

 

[Seaprobe]

"Cleaning Up," comparing what is in the network disk image versus what is on your Mac, and updating each of the changed files and their metadata all create a LOT of network activity.

Much of time machine's activity is updating the directory and metadata structure of the virtual disk image. Once that becomes too big to fit in cache (not just RAM, but cached) it requires a large number of physical disk accesses. Adding RAM keeps most of the activity in fast cache rather than churning with slower physical disk access.

 

[Andy]

Good stuff..so even though the Resource Monitor is not showing the RAM being taxed, I would still benefit from more RAM? i.e those Cleaning Up tasks would utilise the extra RAM

 

[fredbert]

This sounds slow and maybe it's due to WiFi. I use a DS215j on DSM 7 for Time Machine backing up two wired Macs and a wireless MacBook Air: all have SSD storage. I don't experience this slowness.

Each Mac has its own DSM account and they have storage quotas that are around 1.5x their Mac's internal storage.

My SMB advanced settings (focusing on just the enabled ones), also I don't have Transfer Log enabled:

And the Advanced setting for Bonjour (I've one shared folder selected for TM):

 

[Andy]

Thanks fredbert for posting those settings..mine are all the same.

I think I might have found the problem- I set the Quota too low. The Macbook in question has a hard drive of 240Gb and I had set the Quota at 250Gb. The used amount on the drive was only around 160GB so I thought I might get away with 240Gb but maybe not.
Ive changed it to 350Gb now, so I will see if it makes a difference. Will the change in quota happen immediately I wonder.

Mmm no difference.

I think I have to get off Wifi and try a cable. It's a Ubiquiti WAP though so I thought it would be no problem. My download speed is 35 Mbps and my upload speed is 10Mbps- but that's using speetest.net i.e an internet speed checker..is that relevant. I'm confused!

 

[fredbert]

Using speedtest.net will test the speed between your device running the test and a server on the Internet, meaning it will be throttled by the slowest part of the part and that's 90% likely to be your ISP's Internet connection. You can test the LAN-only connection to the NAS by installing iperf3 in a Docker container and getting a client app with iperf3 (or use Homebrew to install on Mac).

I use the networkstatic/iperf3:latest Docker image and either WiFiPerf (from free in Mac App Store) or command line iperf3 is SSH. And only leave the container running while doing a speed test, no need to have ports open on the NAS when not needed.

 

[Andy]

Thanks fredbert..

 

[Andy]

After further investigation: this is not a speed issue so much as a 'jamming' issue. Every backup is super fast (2 minutes) for the first 120Mb or so and then just freezes to a snails pace at around the 120-150Mb mark and subsequently takes several hours to do the remaining 120-150Mb or so.
I cannot think what is causing this!
Any ideas?

 

[fredbert]

I just looked and the MBA and was saying 3 hrs to complete remaining backup (about 700MB). I've just updated SMB Service to see if that does anything. I'm not yet on DSM 7.0.1.

I may bin one of the TM backups and start again to see if that changes something. I only do TM because it's there but I have other backups.

 

[Andy]

Yes, I might ditch and start again. The strange thing is, I only started this recently and the very first (large- 200Gb) back up went really quickly with no problems.

 

[fredbert]

It used to work quickly, well relatively quickly. My DS215j was the first of the NAS to go to DSM 7 and that some months ago. But I hadn't seen any long backups until you mentioned it.

 

[Andy]

OK I've deleted that User and backup Folder and will start again. I have 3 Macs using TM onto this NAS and only one of them is exhibiting this problem, so I've deleted it and will try again and will report back

EDIT: new backup 12Gb in about 20 minutes....but that is as it was the previous initial backup.

 

[Andy]

OK a (hopefully) last report: I deleted the User and the Folder and started from scratch as I said, and all is good! Normal backup times post initial backup on both WiFi and wired.
Not sure what that was all about..

 

[Robbie]

Occasionally Time Machine can get perturbed by a macOS update and ends up reading back massive amounts of data from the NAS (green sawtooth pattern on Rivendell below) on what should be a very small hourly increment. I've not had this happen for ages but caught it doing so recently between my Mac mini and my main NAS.

Bearing in mind this is over a 10GbE link you can see below that the amount of data flowing from one hour's backup almost runs into the next hourly increment and almost all the data is flowing from the NAS to the Mac:

In the above screenshot you can see that I deleted the old backup and set a fresh one running at around 1200hrs (the green spike) with a smaller (purple) spike of backup validation at 1300hrs. From there the scaling makes it look like little is happening from then on but those small blips covers multiple macOS and iOS clients being backed up from then on, every hour.

Almost 2 weeks later my macOS and iOS clients backup schedule now looks like this:

The NAS is now showing a fraction of the GB exchanged than before the fresh TM backup was triggered. Fresh starts can be good and if you prioritise the macOS client by giving it full resources it does not take long at all to complete.

️

 

[fredbert]

I have noticed that TM backup sessions do run high bandwidths back to the Macs. Hadn't worked out why, but TM does keep snapshots on macOS (rather than just the cached data that was always done on MacBooks) when there's spare space to do it ... so housekeeping when I'd rather it was just a versioned backup on the destination. I normally revert back to CCC cloned data that wade through TM.

 

[Andy]

Interesting stuff...

 

[fredbert]

My TM sessions look to have settled down today and haven't seen the estimated hours to complete. I still saw bandwidths of 400Mbps leaving the NAS and entering one Mac, and this NAS only does TM.

I use PeakHour 4 on Mac to monitor interface speeds using SNMP polling.

To allow SNMP v3 polling you need to add a user on the Mac (it's in Control Panel in DSM and SRM). These are the commands I've used, works up to Catalina and I guess nothings changed in Big Sur +, just change the [auth key], [privacy key], and [security name] (same as SNMP community in v1 and 2) which is what PeakHour calls these variables.

sudo net-snmp-config --create-snmpv3-user -ro -a [auth key] -x [privacy key] -X AES -A SHA [security name]

sudo launchctl unload -w /System/Library/LaunchDaemons/org.net-snmp.snmpd.plist
sudo launchctl load -w /System/Library/LaunchDaemons/org.net-snmp.snmpd.plist