ssd swap - high usage of Terabytes Written

[vcsyc]

I did a small test. 2x times.
I left Mac to on battery and closed the lid, then left it for 30 minutes. Then Checked the TBW.
Then I did the same, only on AC power.
I repeated this cycle 2 times and did two row of tests.

For each cycle now Mac had 3x-4x higher TBW when on AC.
I actually did this test twice. On the first try I had only Apple Silicon apps running. I had swap about 1GB.
For the second try I added some Intel apps. Swap increased to 2GB,
but overall results were the same. Mac was writing 3-4 times the data in sleep mode on AC power.

I already did countless times in this topic.

• Swapping doesn't generate enough SSD load to cause 100+ TBW within a couple weeks like some users report.
• If Swapping was the cause this phenomenon would be observable for a long time already, which is not the case.
• If Swapping was the cause, people who only run light tasks such as web browsing would not be affected, yet they are.
• If Swapping was the cause, I would have the issue myself because I am using my 16GB M1 MBP for workloads that have been maxing out my Hackintoshes 32GB of RAM and I am Swapping a lot. Yet here I am at 2.95 TBW after 3.5 months of heavy usage.

Did you leave Mac on sleep? Or maybe you log out or use it only on AC?

 

[VitoBotta]

Just disabled my powernap per terminal:
"sudo pmset -a powernap 0"

Somehow i think it's powernap waking up the M1, checking for updates, then writing RAM back to swap, and that over and over again, multiple times a day...

PS: Once you change any energy settings, it sets powernap back to 1.

Looks like powernap is no longer a thing with Apple Silicon? https://alkhaleejtoday.co/technology/5479664/Mac-M1-computers-that-never-sleep.html

Also it seems that the pmset should show "darkwakes" if it is supported. https://apple.stackexchange.com/que...-os-xs-power-nap-feature-from-within-terminal

 

[VitoBotta]

For those curious, here's a chart of GB increment over a total time of 49h 18m. In the beginning I was checking the SMART data a lot more frequently so the increment is smaller. 1.3+ TB written in those 49 hours. The X axis is the difference in time between the current measurement and the previous time, while the Y axis is the total GB written at each measurement. I must add that this afternoon I enabled Spotlight again after watching the video shared earlier, because I want to use the computer normally. Still, 1.3+ TB in 49 hours...

 

[BacioiuC]

Mac was writing 3-4 times the data in sleep mode on AC power.

I observed and documented the same behaviors on page 26/27 of this thread.

There's something about AC power that wakes up the mac in the background, loads stuff into memory again (and swaps them if not enough memory) and then goes back to sleep only to repeat the cycle.

Since my tests, I disabled sleep again and I only charge my macbook when it's turned off. I think my test ended up with 30-40 gb written in half an hour. Since ~2 days ago, without sleep and charging when off, I only had 12 gb written to my SSD.

 

[Dockland]

I observed and documented the same behaviors on page 26/27 of this thread.

There's something about AC power that wakes up the mac in the background, loads stuff into memory again (and swaps them if not enough memory) and then goes back to sleep only to repeat the cycle.

Since my tests, I disabled sleep again and I only charge my macbook when it's turned off. I think my test ended up with 30-40 gb written in half an hour. Since ~2 days ago, without sleep and charging when off, I only had 12 gb written to my SSD.

I could not help but compare with the meaning of the word Gremlin (off topic i know)

 

[Ningj]

I've had my 1TB 16G Mac Mini since December 3 and set it up from scratch rather than restoring or using migration assistant.

Personal use - web browsing (safari), social/messaging apps, o365 and a number of music apps (Spotify, Garage Band, Ableton Live, Guitarpro). I also use iCloud (40k small files) and spent a lot of time evaluating OneDrive and rsync'ing cloud docs before reverting due to v.high app memory and CPU load.

Time machine is enabled, all preview o365 channel updates applied. System has been completely rebuilt from scratch once, and have applied all Big Sur service packs, GA upgrades, and public beta releases - currently running 11.3 Beta 3. Had an mds/mdsync issue with 11.3 Beta 2 - high CPU due to TM, no unusual write activity noted though.

TBW behaviour looks explainable so far based on usage, maybe a little higher than expected and have started recording it out of interest (5 data points from feb 18 - mar 7). The huge TBW figures reported by some users is concerning and must be down to either the environment or applications. As my profile appears to be reasonably consistent across 11.1 - 11.3 Beta 3 releases, more detail about workloads and apps would be useful.

I'm not using Lightroom as I have yet to decide if I will resubscribe or move to an alternate workflow. Heres my 12mth forecast

 

[badsimian]

I have nearly 16TBW on a 16GB M1 MBA, light use only, mostly browser, terminal stuff. The odd launch of Parallels with Windows on Arm but not often.

 

[Wildkraut]

Looks like powernap is no longer a thing with Apple Silicon? https://alkhaleejtoday.co/technology/5479664/Mac-M1-computers-that-never-sleep.html

Also it seems that the pmset should show "darkwakes" if it is supported. https://apple.stackexchange.com/que...-os-xs-power-nap-feature-from-within-terminal

Hmmm ccording to the stackexchange source link, powernap and darkwakes does the same now:

} else if(0 == strncmp(argv, ARG_DARKWAKES, kMaxArgStringLength) ||
(0 == strncmp(argv, ARG_POWERNAP, kMaxArgStringLength)))
{

FYI, || means “Or”

Anyway, if I tweak anything in the energy panel, it auto sets powernap back to 1. Time will tell, if tweaking powernap helps...

 

[VitoBotta]

Hmmm ccording to the stackexchange source link, powernap and darkwakes does the same now:

} else if(0 == strncmp(argv, ARG_DARKWAKES, kMaxArgStringLength) ||
(0 == strncmp(argv, ARG_POWERNAP, kMaxArgStringLength)))
{

FYI, || means “Or”

Anyway, if I tweak anything in the energy panel, it auto sets powernap back to 1. Time will tell, if tweaking powernap helps...

How do you know if powernap is enabled or not? With pmset I don't see darkwakes.

 

[Wildkraut]

How do you know if powernap is enabled or not? With pmset I don't see darkwakes.

I see “powernap” 1, and both does the same.

 

[VitoBotta]

I see “powernap” 1, and both does the same.

It doesn't show up for me:

❯ pmset -g
System-wide power settings:
Currently in use:
 Sleep On Power Button 1
 autorestart          0
 networkoversleep     0
 disksleep            10
 sleep                10 (sleep prevented by powerd, Amphetamine, mdsync)
 ttyskeepawake        1
 displaysleep         10 (display sleep prevented by Amphetamine)
 womp                 1

However it seems that there have been 2 darkwakes since last night:

❯ pmset -g stats
Sleep Count:3
Dark Wake Count:2
User Wake Count:1

Which Mac do you have? I have the mini.

 

[majormike]

What app would you recommend to track the writing of the SSD in detail? Maybe even based on days? What App would you recommend to track it?

I'm getting my macbook air tomorrow but I'm afraid to use it now

 

[majormike]

I mean, even 8 gb machines shouldn't have such an insane ram to ssd swap while doing nothing, there is obviously something wrong based on the reports. It's not like it can be traced back using a photo, video or audio App which would fulfill that demand.

 

[Wildkraut]

It doesn't show up for me:

❯ pmset -g
System-wide power settings:
Currently in use:
Sleep On Power Button 1
autorestart          0
networkoversleep     0
disksleep            10
sleep                10 (sleep prevented by powerd, Amphetamine, mdsync)
ttyskeepawake        1
displaysleep         10 (display sleep prevented by Amphetamine)
womp                 1

However it seems that there have been 2 darkwakes since last night:

❯ pmset -g stats
Sleep Count:3
Dark Wake Count:2
User Wake Count:1

Which Mac do you have? I have the mini.

I have the MacBook Air 16GB.

powernap is 0, because i've set it to 0, but it was 1.

 

[ManicMarc]

Has anyone noticed any correlation in screen resolution and swap usage? The reason I ask is, I have an Intel Air that was set to 'Default for display" and was using hardly any swap. I moved the display setting up up a notch to show more on screen and swap usage went up. Makes sense as more on screen will equal more shared memory going to graphics (I know the number of pixels is the same, but assume that more content = more memory).
I imagine an 8GB machine with the 6K XDR display would have little left for apps.

 

[Maximara]

I mean, even 8 gb machines shouldn't have such an insane ram to ssd swap while doing nothing, there is obviously something wrong based on the reports. It's not like it can be traced back using a photo, video or audio App which would fulfill that demand.

The problem is that other are reporting no problem at all and now the idea of graphic settings being a possible source is being suggested (which given the unified memory makes perfect sense). Could be Apple's code or something third party app goes off its nut.

 

[VitoBotta]

I have the MacBook Air 16GB.

powernap is 0, because i've set it to 0, but it was 1.

OK, I guess it's not a Mini thing then.

 

[Donga120]

To be fair, that's okay-ish. On my iMac Pro bought at the end of March Last year, I have about 11 TB written to the SSD. 11.92 TB. I do use it for really heavy tasks but it's also a 32GB machine in terms of ram so it's much less swap usage on it.

1TB a month should be okay, it means on average around 30-ish gigabytes / day. For an 8GB ram machine, it's at the higher end of the ballpark. My M1 MBP averages 10 GB written day.

I'm not worried about 1TB a month, it is a lot but the machine can take it, some guys on here are writing that in a few days

 

[Never mind]

Another video you may wish to look at....

 

[dsusanj]

I have just had my M1 MBP (8/256) write over 1 TB in an hour by the kernel_task.

The cause was easy to pinpoint: Parallels Technical Preview running Windows for Arm.

As soon as Parallels is launched (with Windows suspended i.e. not even running!), writes skyrocket from 150 KB/s to 300-400 MB/s (!) (sometimes over 500 MB/s) and stay that way as long as Parallels is running, having in the end written over 1 TB to the SSD in an hour. As soon as Parallels is shut down, writes go back to low levels.

 

[gank41]

I have just had my M1 MBP (8/256) write over 1 TB in an hour by the kernel_task.

The cause was easy to pinpoint: Parallels Technical Preview running Windows for Arm.

As soon as Parallels is launched (with Windows suspended i.e. not even running!), writes skyrocket from 150 KB/s to 300-400 MB/s (!) and stay that way as long as Parallels is running, having in the end written over 1 TB to the SSD in an hour.

I use the Parallels Tech Preview w/ both Windows 10 for ARM and Ubuntu, and am not seeing that happen on my end.

 

[Forti]

Some more details here:

 

[Baff]

I'm not worried about 1TB a month, it is a lot but the machine can take it, some guys on here are writing that in a few days

Yeah, I would not be too worried about 1TB/month. My Intel Mini averaged 1.25TB/month over 6 years and is chugging along fine. That was with 70-100 tabs open in Chrome, plus a variety of other apps always open, and a fair amount of gaming at times.

I have managed to get the writes on my M1 Mini under control by closing the biggest memory hog tabs in Safari. My swap is down to 0 right now and the only writes so far today are due to me streaming some Netflix. Yesterday my swap was at 150-700MB most of the day and it wrote about 30GB, not great but acceptable. 2 days ago my swap was at 5-7GB and it wrote 400GB, which is horrible.

I really hope Apple figures this out and fixes it, I really don't want to spend years watching Activity Monitor constantly in order to make sure it doesn't go bonkers.

 

[jdb8167]

In order to do their survey it requires a Google login. Yeah that’s not happening. What a stupid thing to require.

 

[Forti]

In order to do their survey it requires a Google login. Yeah that’s not happening. What a stupid thing to require.

And what exactly can they do with your 1-hour fake Google account that you can create within 5 mins? ;p