ssd swap - high usage of Terabytes Written

[leman]

To provide a reference, here is my vm_stat after nearly 2 days of uptime:

Mach Virtual Memory Statistics: (page size of 16384 bytes)
Pages free:                                6013.
Pages active:                            846388.
Pages inactive:                          807221.
Pages speculative:                        36494.
Pages throttled:                              0.
Pages wired down:                        154665.
Pages purgeable:                          15069.
"Translation faults":                 400321787.
Pages copy-on-write:                    6903167.
Pages zero filled:                    151347241.
Pages reactivated:                      1425949.
Pages purged:                           1331405.
File-backed pages:                       629193.
Anonymous pages:                        1060910.
Pages stored in compressor:              501711.
Pages occupied by compressor:            192899.
Decompressions:                          321359.
Compressions:                           1005651.
Pageins:                                2460036.
Pageouts:                                 37586.
Swapins:                                   4416.
Swapouts:                                  4576.

The swap outs are about 75MB or 71.5 MiB,

Thanks, does your machine has 8GB oder 16GB?

However, the data written since reboot is 1000 times more (71GB).

View attachment 1987768

I haven't done any large file copies, so that it quite a discrepancy!

Confirms the common sense logic that swap data transfer should be negligible in the grand scheme of things.

What is the difference between Pageouts/ins and Swapouts/ins?

It’s impossible to answer properly without doing an in-depth study of the Darwin pager code. But in general not every page out event results in data being written to the swap. Things like memory compression, dropping purgeable memory or simply discarding data that has existing backing on the disk are also page outs.

EDIT: looked at some numbers again and now I believe that "pageouts" primarily refers to discarding memory pages already backed by some data on the disk.

 

[ctjack]

When did you last reboot or shut down your computer?

7 days 2 hours.
Other stats:

Spoiler: pics

 

[Queen6]

If you're worried about SSD wear, spend a little on the next tier of memory or storage. It's designed like that.

Base M1 Macs are meant for light productivity and browsing.

LOL, he/she clearly isn't neither am I. People purchase what they want and use as they see fit, be it an Air in a professional development role or a top tier MBP to surf the web simply because it looks nice on the coffee table.

If this was such a serious issue it would have blown up by now given the M1 Mac's have been out in the wild for a good time. Vast majority use their Mac's casually, professionals are either provided with their hardware as employees or directly capitalise on the hardware which in turn pays for itself.

Personally I buy Mac's to use both casually & professionally I don't expect them to last in perpetude. I just expect them to get the job done be it a base model or a high tier MBP for a reasonable term. As for the base M1 Mac's they remain tremendously capable easily out performing far larger notebooks. The M1 took the computing world by a storm on release and last I checked they haven't gotten any slower and the rest just got better for those that have the need or simply just like...

Q-6

 

[leman]

7 days 2 hours.
Other stats:

Spoiler: pics

View attachment 1987946 View attachment 1987947

Yeah, that's over 100GB of swap written per day. Definitely sounds excessive. But looking at your system it's clearly memory starved. Maybe I should readjust my RAM estimates after all…

 

[jdb8167]

Because we don't see these SSDs failing en masse. M1 Macs have been out for almost two years now, and a lot of folks own the base 8GB/256GB model. If high SSD writes were a common thing we would start seeing Airs failing left and right about half a year ago. But it did not happen.

We aren't even close to 2 years yet—less than 18 months even. I don't know why people keep saying we are near the 2 year mark. The first release was near the end of November 2020.

I haven't seen anyone report more than about 80% used. Those are probably the high mark or close to it. I expect a wave of failures around Spring 2023 with the 8GB/256GB M1 Macs. The 256 GB makes the situation worse in that not only is the TBW lower but I'd bet that most 256 GB SSDs are significantly filled and that makes write amplification worse than average as well. Of course it depends on if Apple over provisioned the SSDs but I think that is pretty unlikely.

 

[ctjack]

We aren't even close to 2 years yet—less than 18 months even. I don't know why people keep saying we are near the 2 year mark. The first release was near the end of November 2020.

I haven't seen anyone report more than about 80% used. Those are probably the high mark or close to it. I expect a wave of failures around Spring 2023 with the 8GB/256GB M1 Macs.

As it was with a flexgate. Apple made a shorter cable that should rub itself down after the warranty mark. Unluckily for them, class lawsuit action made them fix it on May 2019 - 3 years after issuing the first failure in MBP 2016.
The same awaits SSDs in these Macs, someday those PB should create a failure. However not sure if Apple will admit it this time.
My gut feeling is that they tested the screen cable long enough to predict a failure after 3 years, which would be good for Apple - no warranty, no Apple care, no UK 2 year warranty. You have to buy a new laptop and say that things happen. I bet some engineers lost their job at the testing and prediction site, because they started failing after 1 year - those who close/open the lid multiple times made things get worse faster than it was predicted.

 

[jdb8167]

As it was with a flexgate. Apple made a shorter cable that should rub itself down after the warranty mark. Unluckily for them, class lawsuit action made them fix it on May 2019 - 3 years after issuing the first failure in MBP 2016.
The same awaits SSDs in these Macs, someday those PB should create a failure. However not sure if Apple will admit it this time.
My gut feeling is that they tested the screen cable long enough to predict a failure after 3 years, which would be good for Apple - no warranty, no Apple care, no UK 2 year warranty. You have to buy a new laptop and say that things happen. I bet some engineers lost their job at the testing and prediction site, because they started failing after 1 year - those who close/open the lid multiple times made things get worse faster than it was predicted.
View attachment 1988083

I doubt either of these problems are deliberate. Long term testing can be difficult since you have to simulate wear. Obviously you can't test a product for 3 years.

The biggest problem with the high SSD write problem is that Apple is certainly aware of it and doesn't seem to be doing anything about it in particular. Maybe there is no ready solution or maybe they can predict the total number of failures and did a calculation that it would be small enough that they can handle it without making expensive wholesale changes to macOS. I suspect the number of people who will have problems is still going to small compared to the total installed base of M1 Macs. It will likely only affect people with 8GB/256GB M1 Macs who do an unusual amount of work on an entry level machine.

I have a 16 GB/1 TB and my current TBW is about 32 TBW after 16.5 months.

 

[ctjack]

It will likely only affect people with 8GB/256GB M1 Macs who do an unusual amount of work on an entry level machine.

I have a 16 GB/1 TB and my current TBW is about 32 TBW after 16.5 months.

So far some people found out that the numbers stop growing exponentially if one is not using Rosetta apps and Spotlight find feature. The less we press cmd+space to look-up a file/software, the less will be the writes number.

 

[jdb8167]

So far some people found out that the numbers stop growing exponentially if one is not using Rosetta apps and Spotlight find feature. The less we press cmd+space to look-up a file/software, the less will be the writes number.

There is a memory leak with Finder and Spotlight. You can use option-right click on the Finder icon in the dock and click Relaunch and the memory is temporarily reset. I don't know any reason why Rosetta 2 should give unusually high number of writes. It's more likely that some applications that have been translated by Rosetta is causing a problem rather than all Rosetta 2 apps.

 

[Queen6]

We aren't even close to 2 years yet—less than 18 months even. I don't know why people keep saying we are near the 2 year mark. The first release was near the end of November 2020.

I haven't seen anyone report more than about 80% used. Those are probably the high mark or close to it. I expect a wave of failures around Spring 2023 with the 8GB/256GB M1 Macs. The 256 GB makes the situation worse in that not only is the TBW lower but I'd bet that most 256 GB SSDs are significantly filled and that makes write amplification worse than average as well. Of course it depends on if Apple over provisioned the SSDs but I think that is pretty unlikely.

If so Apple will be hit by multiple class action law suits, as lawyers, money & opportunity will all align LOL. In the meantime I've an auto destruct 15"MBP 2011 that refuses to die. I kind of suspect the vast majority of the Apple Silicon Mac's will be siting with my 2011 MBP; functional yet outdated.

Yes some will burn out the SSD; here is a forum for enthusiasts. Companies tend to rotate out or repurpose computers for multiple reasons. Individual users don't push the hardware anywhere near close to the same extremes. Then we have the outliers who break all the rules and push their systems to the max. Similar to a sharp knife it can blunt with use...

Q-6

 

[Fomalhaut]

Thanks, does your machine has 8GB oder 16GB?



Confirms the common sense logic that swap data transfer should be negligible in the grand scheme of things.



It’s impossible to answer properly without doing an in-depth study of the Darwin pager code. But in general not every page out event results in data being written to the swap. Things like memory compression, dropping purgeable memory or simply discarding data that has existing backing on the disk are also page outs.

EDIT: looked at some numbers again and now I believe that "pageouts" primarily refers to discarding memory pages already backed by some data on the disk.

Thanks. So it's still unclear what actually writes to the disk. I know I can run a tool that show the disk writes for all processes (I did this last year when this thread started), but it was a non-trivial to interpret!

BTW, my machine is a 32GB MBP14 Max.

 

[Maximara]

We aren't even close to 2 years yet—less than 18 months even. I don't know why people keep saying we are near the 2 year mark. The first release was near the end of November 2020.

The first Macs with SSDs options, rather than fusion drives, were January 2008 Macbook Air with exclusively SSD models being late-2010 Macbook Air, mid-2012 ‘Retina’ Macbook Pro. Some time ago I stated this:

Percentage Used: 100%, Available Spare Threshold: 10%; 448 TBW on 5-year old MBP. Yes, he didn't state the size of the drive but as you can see this is not unique to Big Sur or M1 Macs. Neither existed in 2016-2017 but rather Intel and Sierra and High Sierra did. So this supposed "write issue" has been around for a long time.

I haven't seen anyone report more than about 80% used.

Now you do and that Mac AFAWK is still doing fine.

Those are probably the high mark or close to it. I expect a wave of failures around Spring 2023 with the 8GB/256GB M1 Macs. The 256 GB makes the situation worse in that not only is the TBW lower but I'd bet that most 256 GB SSDs are significantly filled and that makes write amplification worse than average as well. Of course it depends on if Apple over provisioned the SSDs but I think that is pretty unlikely.

Given how uncommon this issue is I don't expect a "wave of failures" though I will say that filling a 256 GB SSD rather just putting the system and the bare minimum of programs that must be on the internal drive and the rest on an external SSD just being shortsighted. The whole I' don't want to have a dongle' response this solution causes is IMHO a really silly and narrow minded response to this really basic solution to what from what I has seen is mostly a Nothing Burger.

Nevermind as I have stated (and some have demonstrated here on Macrumors) third party programs like Teams, anything using Electron/Chromium, and anything with a large amount of x86 code has been identified as SSD writers even with high amounts of free RAM. Apple has no control over those programs or users doing something that kills their Mac. By contrast the fiasco that Monterey 12.3 was is something that is totally Apple's fault.

 

[jdb8167]

The first Macs with SSDs options, rather than fusion drives, were January 2008 Macbook Air with exclusively SSD models being late-2010 Macbook Air, mid-2012 ‘Retina’ Macbook Pro.

None of this is relevant to the Apple silicon SSDs. Try to stay on topic.

 

[Maximara]

Thanks. So it's still unclear what actually writes to the disk. I know I can run a tool that show the disk writes for all processes (I did this last year when this thread started), but it was a non-trivial to interpret!

BTW, my machine is a 32GB MBP14 Max.

Well Activity Monitor will provide some information (only from the last boot up). The problem is that as explained back in a 2017 thread: "kernel_task is a core part of the OS that is sort of like the traffic police for many different processes. So apps that write to the disk do it through kernel_task." So heavy writes by kernel_task do not always mean heavy Swap.

None of this is relevant to the Apple silicon SSDs. Try to stay on topic.

The topic is "ssd swap - high usage of Terabytes Written" which has nothing in the title regarding M1 in it. Also from some of the early posts (Feb 2021) also used Intel Macs:

* "I checked on my MacBook Pro 13" from 2018, 256/16 as well. I used both machines for the same task (99% of time it's coding, swift, javascript etc)"
* Mac Pro Big Sur 10 months - ~ 67TB; that would have been an Intel Mac as there were no M1 Macs available in April 2020
* "1yo MBP16 (fairly heavy torrent user):
Data Units Read: 66,762,250 [34.1 TB]
Data Units Written: 27,272,874 [13.9 TB]"

Ergo what I posted is "on topic".

Again if it is a software issue it would be effecting everyone and even if it was only M1 hardware (for what ever weird reason) then everyone who uses the SSD drive check tool would se it and some are not seeing it. Per Occam's razor it is third party programs that are mainly causing the issue.

In fact I found the post (by jdb8167 on Feb 24, 2021) regarding the "references" smartmontools uses:

"How has it been proven? I haven't seen anything definitive. I've looked at the smartmontools code and even their code documentation says they are guessing and they only have two specific drives in their database neither of which matches the NVMe SSD in the M1 Macs."

 

[jdb8167]

The topic is "ssd swap - high usage of Terabytes Written" which has nothing in the title regarding M1 in it. Also from some of the early posts (Feb 2021) also used Intel Macs:

Look at the forum please. You are off topic.

 

[jdb8167]

In fact I found the post (by jdb8167 on Feb 24, 2021) regarding the "references" smartmontools uses:

"How has it been proven? I haven't seen anything definitive. I've looked at the smartmontools code and even their code documentation says they are guessing and they only have two specific drives in their database neither of which matches the NVMe SSD in the M1 Macs."

I retracted that later as I did more research into their open source codebase. I was completely wrong in that post.

Edit: In case it is unclear. I gradually educated myself on the smartmontools. Here is a later post of mine on this topic.
https://forums.macrumors.com/threads/ssd-swap-high-usage-of-terabytes-written.2284893/post-29736750

 

[Maximara]

But why would that mattered to the point to even note it in the documentation?

 

[jdb8167]

But why would that mattered to the point to even note it in the documentation?

The code is old and comments don’t always reflect current knowledge. Ask any developer, it is completely normal if a bit unfortunate.

 

[darngooddesign]

The first Macs with SSDs options, rather than fusion drives, were January 2008 Macbook Air with exclusively SSD models being late-2010 Macbook Air, mid-2012 ‘Retina’ Macbook Pro.

I thought the original MBAs had tiny and slow platter drives and the SSDs came in 2010 with the wedge redesign.

 

[Maximara]

I thought the original MBAs had tiny and slow platter drives and the SSDs came in 2010 with the wedge redesign.

As I said it was an option for the January 2008 Macbook Air not standard. And it was not cheap either:

"This January marks the 10th anniversary of the iconic MacBook Air. Presented below is our original review of the first-generation Air from January 24, 2008. It wowed us with its design, but also struck us as less than practical for most. The most notable surprises about this initial incarnation of the MacBook Air were its single USB-A port, lack of an optical drive, and the budget-busting $1,799 starting price (plus an extra $1,000 to upgrade to SSD storage). " (...) "While the 80GB hard drive included in the base $1,799 model may be smaller than you're used to, the only other option is a 64GB solid state hard drive." - Apple MacBook Air review:MacBook Air at 10: A look back at our original 2008 review

"The future may have to wait a few years for prices to come down; however, swapping the 80GB platter drive for the 64GB SSD drive is a whopping $999 upgrade. The only other internal hardware option is a CPU uptick, from 1.6GHz to 1.8GHz for $300. With the upgraded CPU and SSD drive, the $1,799 MacBook Air suddenly becomes a $3,098 laptop." - MacBook Air revisited, CNet, Feb 4, 2008

 

[evertjr]

There is a memory leak with Finder and Spotlight. You can use option-right click on the Finder icon in the dock and click Relaunch and the memory is temporarily reset. I don't know any reason why Rosetta 2 should give unusually high number of writes. It's more likely that some applications that have been translated by Rosetta is causing a problem rather than all Rosetta 2 apps.

Thank you for pointing that out, I'll occasionally relaunch Finder. Now I think there's some software issue trashing the ssd. Yesterday morning the only thing different I did was to use Grand Perpective to do some cleaning and deleted some kind of dlyb cache related to iOS simulators that was taking some space. Its probably unrelated but surprisingly that same day my average disk writes reduced drastically from 1TB on the day before to 100gb with the same intense usage (over 5gb of swap with MS Teams running all day). Maybe its related to this finder/spotlight issue.

 

[darngooddesign]

As I said it was an option for the January 2008 Macbook Air not standard. And it was not cheap either:

That right. While I forgot that it was an option I didn't forget that it was crazy expensive. $999 for 64GB. LOL

 

[Fomalhaut]

Well Activity Monitor will provide some information (only from the last boot up). The problem is that as explained back in a 2017 thread: "kernel_task is a core part of the OS that is sort of like the traffic police for many different processes. So apps that write to the disk do it through kernel_task." So heavy writes by kernel_task do not always mean heavy Swap.

View attachment 1988397

I hadn't thought to look at "Bytes Written" per process, and it's interesting...

As expected "kernel_task" and "launchd" were the largest writers. I think "fileproviderd" comes from one of my cloud storage tools (Google Drive, Dropbox or OneDrive) and "mds_stores" may be related to indexing or Time Machine? But it was surprising to see that iStatMenus and Safari Bookmarks Sync write so much...no idea what they are writing...

Just goes to show that there is a lot of disk writing from apps and processes that you wouldn't think need to write much data.

 

[Mike Boreham]

But it was surprising to see that iStatMenus and Safari Bookmarks Sync write so much...no idea what they are writing...

iStat Menus saves all data for 30 days, which is very useful for "after the event" troubleshooting as I found after my big TBW event.

 

[qwezxc]

is this still a problem ? I'm new to Mac OS , have a m1 pro 14 32 gb 1tb ssd with this data,

This seems way too much :

SMART/Health Information (NVMe Log 0x02)

Critical Warning: 0x00

Temperature: 34 Celsius

Available Spare: 100%

Available Spare Threshold: 99%

Percentage Used: 0%

Data Units Read: 3 513 782 [1,79 TB]

Data Units Written: 3 182 023 [1,62 TB]

Host Read Commands: 53 246 747

Host Write Commands: 59 603 916

Controller Busy Time: 0

Power Cycles: 123

Power On Hours: 39

Unsafe Shutdowns: 5

Media and Data Integrity Errors: 0

Error Information Log Entries: 0