ssd swap - high usage of Terabytes Written

[slvrscoobie]

since my last report about a week ago, I went from 3.9 to 4.4tb - guess im doing something right

 

[dogslobber]

Sorry to rain in your parade but Alfred DOES NEED spotlight to work. From Alfred's web:
[...]
In any case, I see your point. We all want Apple to fix this thing. Moreover, several people have tried turning Spotlight indexing off and the problem still persists.

Try QS then if Alfred doesn't work. I'm not sure how people are "turning Spotlight indexing off" but I posted previously how to do it and it's been effective for me.

 

[majormike]

It's hilarious how everybody in the Apple user base has been okay with soldered on SSDs, went on recommending only 8 Gigabytes ram to be sufficient for the M1 Macbooks in most cases while these machines heavily use the system storage to cache for insufficient ram, creating several terabytes of written data after only a few days 🤣 I'd call this the perfect soap drop and bendover in Apple's own history.

A simple swappable SSD slot would've resolved that issue, but Apple isn't too much into making their computers reliable and long lasting anymore.

 

[flopticalcube]

It's hilarious how everybody in the Apple user base has been okay with soldered on SSDs, went on recommending only 8 Gigabytes ram to be sufficient for the M1 Macbooks in most cases while these machines heavily use the system storage to cache for insufficient ram, creating several terabytes of written data after only a few days 🤣 I'd call this the perfect soap drop and bendover in Apple's own history.

A simple swappable SSD slot would've resolved that issue, but Apple isn't too much into making their computers reliable and long lasting anymore.

Swap hasn't been identified as the issue yet. The SSD is no longer replaceable because the NVME controller is part of the M1 SoC. Only the flash memory remains external. Apple could have put that on a daughter-board but that would be proprietary and possibly entail some engineering tradeoffs and add cost.

 

[alien3dx]

It's hilarious how everybody in the Apple user base has been okay with soldered on SSDs, went on recommending only 8 Gigabytes ram to be sufficient for the M1 Macbooks in most cases while these machines heavily use the system storage to cache for insufficient ram, creating several terabytes of written data after only a few days 🤣 I'd call this the perfect soap drop and bendover in Apple's own history.

A simple swappable SSD slot would've resolved that issue, but Apple isn't too much into making their computers reliable and long lasting anymore.

haha.. it will be switchable like Microsoft said ..
"Microsoft Ready Boost" -> for low ram user . Normally we can set swap drive in linux or windows easily and i'm not sure it can be done in macOS.

 

[rui no onna]

It's hilarious how everybody in the Apple user base has been okay with soldered on SSDs, went on recommending only 8 Gigabytes ram to be sufficient for the M1 Macbooks in most cases while these machines heavily use the system storage to cache for insufficient ram, creating several terabytes of written data after only a few days 🤣 I'd call this the perfect soap drop and bendover in Apple's own history.

A simple swappable SSD slot would've resolved that issue, but Apple isn't too much into making their computers reliable and long lasting anymore.

A swappable SSD doesn't really fix the issue, though. It just masks the symptoms.

 

[ventmore]

If that interests you go ahead. What works for me is to neuter spotlight indexing until Apple fixes this stopper bug.

If I owned an M1, and if it had this issue, then I'd definitely be trying to narrow it down......my curiosity wouldn't let me stop at just disabling spotlight altogether.

 

[RvXtm]

Using Lightroom CC for about one hour to edit 3 raw files. 200GB swap written to SSD.
I can use my external ssd, the segate firecuda, to host the os, but this is not a fix.
I tried to install BigSur on the external segate ssd, same crazy writers on that drive as well. At least i am not hurting the internal drive that hosts the preboot, bootloader and firmware.

 

[qap]

Why are you all saying that is swap? To me it doesn’t look a swap file, I can have 100Gb written on disk but only 2/3gb of swap file. For that I don’t think that resize or disable it will change anything.

To me looks more a bug with the kernel_task that is paging in/out too much!

 

[thenewperson]

A swappable SSD doesn't really fix the issue, though. It just masks the symptoms.

Yeah, that's just allowing you to go through even more SSDs in this way. There's clearly an issue with things being written to disc unnecessarily within the OS (or some rogue app). Replaceable RAM/SSDs won't really help.

 

[Forti]

Why are you all saying that is swap? To me it doesn’t look a swap file, I can have 100Gb written on disk but only 2/3gb of swap file. For that I don’t think that resize or disable it will change anything.

To me looks more a bug with the kernel_task that is paging in/out too much!

well, I'm not so sure if kernel_task isn't responsible for the swap - it's one of the most fundamental task of the system to manage the RAM, right?

And from the observations, reading twitter, this thread, some articles out there in the web - it looks like:

the less RAM you computer have - the more SSD TBW can be created.

 

[qap]

well, I'm not so sure if kernel_task isn't responsible for the swap - it's one of the most fundamental task of the system to manage the RAM, right?

And from the observations, reading twitter, this thread, some articles out there in the web - it looks like:

the less RAM you computer have - the more SSD TBW can be created.

Yes it’s responsible for the swap file but if the swap file is small, it’s writing something else for other reasons.
For example kernel_task can use a lot of “null/fake” cpu cycles in order to cool down the Mac.
To me this looks something similar, where the kernel is paging i/o for something not necessary. That is why is reporting a lot of TB written on disk but the the swap is only few GB (and this is normal).

I already checked my logs but looks all okay 🤔

 

[vcsyc]

Yes it’s responsible for the swap file but if the swap file is small, it’s writing something else for other reasons.
For example kernel_task can use a lot of “null/fake” cpu cycles in order to cool down the Mac.
To me this looks something similar, where the kernel is paging i/o for something not necessary. That is why is reporting a lot of TB written on disk but the the swap is only few GB (and this is normal).

I already checked my logs but looks all okay 🤔

I imagine that SWAP file should be rewritten several times.
so if swap is 1GB, is it one static file or different info in an 1GB allocated memory?
In this case, how many times lets say in an hour does the system does rewrite swap file?

 

[qap]

I imagine that SWAP file should be rewritten several times.
so if swap is 1GB, is it one static file or different info in an 1GB allocated memory?
In this case, how many times lets say in an hour does the system does rewrite swap file?

Swap file is a part of memory that is needed by an app or system but is not requested often or on the last X time.

For example if you open photoshop and you edit a file, it uses 2gb of Ram, then you left PS open but you switch to other tasks/apps that require additional RAM, so the OS should put the 2Gb of PS (or a part of it) in a swap file on the disk because the OS/new apps need the ram. And usually the swap file is not touched/emptied if is not needed for other task (I mean the storage on SSD) or you don’t re-open Photoshop.

So it should not continuously write or read the portion of SSD allocated to swap, it’s a bug/weird behavior.

But for me at least, the swap file has a reasonable size and it doesn’t increase, so the kernel_task is writing something else somewhere. That is the weird thing that is happening on ARM macOS.

 

[ambient_light]

For those in doubt, here's a bit more numbers from VM management in support to swapping bug in the macos kernel:

• kernel_task reported 284 Gb written out through Activity Monitor
• vm_stat reported 17+M pages 16K each swapped in (written to swap file) which gives you roughly 270+Gb
• TBW from Activity Monitor is 336Gb (12 hrs uptime)

That gives 80+% of write activity was caused by swapping, and the laptop was either lightly used or inactive most of those 12hrs. I'd note that swapping activity from vm_stat is much higher than paging, and is almost always on.

Long story short it all screams VM kernel bug to me.

 

[theSeb]

I imagine that SWAP file should be rewritten several times.
so if swap is 1GB, is it one static file or different info in an 1GB allocated memory?
In this case, how many times lets say in an hour does the system does rewrite swap file?

Every time there is a page out this is recorded and can be checked by running vm_stat. This keeps the page outs since the last restart.

M1 Macs do seem to write to swap and page out more readily, but that does only accounts for some extra GBs according to vm_stat and does not correlate to the TBs of data written according to vm_stats shared by other users, apart from ambient_light's post just above, which is very compelling evidence.

If you search around the internet and this forum you will find occasional users having issues with kernel_task writing TBs of data, but these were isolated incidents reported by a few users and were tracked down to specific issues. But this new problem appears to be very widespread on M1 Macs and the total TBWs reported by users here and elsewhere with M1 Macs is obscene.

 

[pistonpilot]

People are not dumb when it comes to their hardware. They can easily look at the write cycles of an Intel Mini running for years and compare it to their 2 month old M1 Mini that has written more in 10x less time. You don't need to be a mathematics professor to spot a problem. Maybe those who don't see a problem aren't very good at maths. I don't know.

They are not good at many things; common-sense being one of them.

 

[majormike]

For those in doubt, here's a bit more numbers from VM management in support to swapping bug in the macos kernel:

• kernel_task reported 284 Gb written out through Activity Monitor
• vm_stat reported 17+M pages 16K each swapped in (written to swap file) which gives you roughly 270+Gb
• TBW from Activity Monitor is 336Gb (12 hrs uptime)

That gives 80+% of write activity was caused by swapping, and the laptop was either lightly used or inactive most of those 12hrs. I'd note that swapping activity from vm_stat is much higher than paging, and is almost always on.

Long story short it all screams VM kernel bug to me.

But then, having your Macbook affected, shouldn't Apple then replace your SSD or send you a new one / to everybody who owns an M1? I'd not be willing to accept that my device has already written a big amount of it's SSDs lifetime.

My new Macbook Air M1 will actually arrive today....

 

[pistonpilot]

It's hilarious how everybody in the Apple user base has been okay with soldered on SSDs, went on recommending only 8 Gigabytes ram to be sufficient for the M1 Macbooks in most cases while these machines heavily use the system storage to cache for insufficient ram, creating several terabytes of written data after only a few days 🤣 I'd call this the perfect soap drop and bendover in Apple's own history.

A simple swappable SSD slot would've resolved that issue, but Apple isn't too much into making their computers reliable and long lasting anymore.

Amen brother. Having had 5 iMac replacements courtesy of AppleCare I can tell you that you are spot on. Apple's hardware sucks. I suspect that both my Intel iMac and Silicon Mac Mini will be replaced, as both have AppleCare and I know the head of USA Customer Relations.

Apple should pull their manufacturing from China immediately. The reliability is sorely lacking.

 

[Goldfinger]

For what it's worth guys, as a data point:
My 4 year (give or take 2 months) old 13" MBP (16GB, 512GB) has the following numbers:
58,7TB read
44,8TB Written
689 power on hours
36 unsafe shutdowns
127805 power cycles
3% life percentage used
84 available spare

 

[theSeb]

For those in doubt, here's a bit more numbers from VM management in support to swapping bug in the macos kernel:

• kernel_task reported 284 Gb written out through Activity Monitor
• vm_stat reported 17+M pages 16K each swapped in (written to swap file) which gives you roughly 270+Gb
• TBW from Activity Monitor is 336Gb (12 hrs uptime)

That gives 80+% of write activity was caused by swapping, and the laptop was either lightly used or inactive most of those 12hrs. I'd note that swapping activity from vm_stat is much higher than paging, and is almost always on.

Long story short it all screams VM kernel bug to me.

This is interesting... the page outs are not significant, but the swap ins are, which are compressed pages written to disk. So the issue is with aggressive memory compression possibly. I also note that pages are 16,384 bytes. On my Intel macs the pages are 4096 bytes each. This could also be a factor causing more writes than are necessary.

 

[theSeb]

Amen brother. Having had 5 iMac replacements courtesy of AppleCare I can tell you that you are spot on. Apple's hardware sucks. I suspect that both my Intel iMac and Silicon Mac Mini will be replaced, as both have AppleCare and I know the head of USA Customer Relations.

Apple should pull their manufacturing from China immediately. The reliability is sorely lacking.

We are discussing a software issue here. The software wasn't "manufactured in China". The design of how the memory works in Big Sur was also not manufactured in China. The problem has nothing to do with hardware.

 

[ambient_light]

But then, having your Macbook affected, shouldn't Apple then replace your SSD or send you a new one / to everybody who owns an M1? I'd not be willing to accept that my device has already written a big amount of it's SSDs lifetime.

My new Macbook Air M1 will actually arrive today....

There were examples in the past when Apple did a widespread HW replacements, e.g. batteries in older iPhone models. The longer Apple releasing the fix, the more widespread problem with wore out SSDs gonna be, and hence the more chances for warranty cases and a class action ... we'll see. Hope it's not something inherently broken in the unified memory architecture and the fix is straightforward.

Before it's solved, the general advice is to reboot the laptop once a day to reset the swap activity, use memory sparingly whenever possible, and shut down background processes which are memory hogs and/or actively writing to the disk (Spotlight, iCloud...).

 

[ambient_light]

This is interesting... the page outs are not significant, but the swap ins are, which are compressed pages written to disk. So the issue is with aggressive memory compression possibly. I also note that pages are 16,384 bytes. On my Intel macs the pages are 4096 bytes each. This could also be a factor causing more writes than are necessary.

Good point, though as far as I can tell it's not just a factor of 4 - kernel_task just keeps writing to the swap continuously even if the memory is available, i.e. the memory management algorithm is broken.

 

[pistonpilot]

We are discussing a software issue here. The software wasn't "manufactured in China". The design of how the memory works in Big Sur was also not manufactured in China. The problem has nothing to do with hardware.

Tell us about your expertise that led you to this conclusion. What are the facts that you used to make this determination? How do you know that there is no hardware flaw that the software is exposing? Do tell us all about your subject matter knowledge that affords you to admonish me in any way.