Overprovisioning?

[haralds]

My last system SSD died after two years of heavy use.

Some reports indicate that overprovisioning has benefits. But it seems like as long as trim is enabled and the drive is left with a lot of free space the same impact is achieved.

I am using dual MX500 2TB for system and data formatted APFS. On Windows the Crucial Storage executive provides a special function to reserve space for OP.

I can see the need for OP designated space without trim.

No clear answer in articles I searched.

 

[justinkr]

Hmm~ never heard of it. And I never had that problem. I am currently running my Mac pro with Samsung SSD 500GB and 2 1tb hard driver were raided 0. Still run never had any issues. Knock on wood.

 

[h9826790]

Hmm~ never heard of it. And I never had that problem. I am currently running my Mac pro with Samsung SSD 500GB and 2 1tb hard driver were raided 0. Still run never had any issues. Knock on wood.

With TRIM ON, all free space are effectively serve as OP. So, no need to reserve more space for OP.

 

[AidenShaw]

With TRIM ON, all free space are effectively serve as OP. So, no need to reserve more space for OP.

To be more precise - all space from files deleted when TRIM is active is effectively OP.

If the disk was ever run with TRIM disabled, free space may or may not be OP. Disk utility can scrub free space and make sure that it is TRIM'd and used as OP.

 

[deconstruct60]

My last system SSD died after two years of heavy use.

Some reports indicate that overprovisioning has benefits. But it seems like as long as trim is enabled and the drive is left with a lot of free space the same impact is achieved.

The latter is important. Cranking up the over provisioning means no end user level application can eat too far into the free space. ( Can easily do the same thing without a fancy tool by just partitioning the drive so that one parturition (e.g. 10%) was left unformatted or used. )

Some folks think that 10% is "a lot of free space" and run their drives done to single digits. It is not.

I am using dual MX500 2TB for system and data formatted APFS. On Windows the Crucial Storage executive provides a special function to reserve space for OP.

Mixing vastly different workloads onto a single disk can have impact also.

I can see the need for OP designated space without trim.

TRIM won't do much if have bulky stuff that sits and one place and the wear leveling doesn't spread it out over the aging cells over time. There so also be time for the SSD to do internal housekeeping. If constantly hammering the driver 24/7 or sending it to sleep whenever not in active "meltdown" high use then the drive firmware may never take the time to do "Spring cleaning". ( Firmware that is expecting a more pedestrian pace may wait for an 'open' window. Firmware that expects to be hammered 24/7 will take steps to weave that in. )



A sanity check though would be to get access to the S.M.A.R.T. data before fail next time and it is really wear or some other condition ( heat , flakey NAND chips , etc.) that is driving the failure. Drive present dead when the drive metadata gets borked also.

 

[haralds]

Thanks. It all makes sense.
Need to break the habit of filling disks to 90%.

 

[AidenShaw]

Thanks. It all makes sense.
Need to break the habit of filling disks to 90%.

And get in the habit of regularly running S.M.A.R.T. or drive vendor tools to check the health of the drive. They often give lots of warning that the drive is having problems - particularly for FLASH wear issues.

Samsung's "Magician" is great for Samsung drives - but it only runs on popular operating systems. On my home workstation it shows

​

for the SSD with the OS partition.

If I click on the "S.M.A.R.T.", I see lots of details about internal error and log counters. They can be frightening - drives get and recover from lots of errors. In particular, that 2 TB Samsung had two bad regions from the first week. They were recovered and replaced from the spare pool - but I still see those two fatal errors in the S.M.A.R.T. data every time I look.

(And if you wonder why a 2 TB drive only has 700 GiB of data - let's just say that I'm into extreme over-provisioning even though TRIM is supported at all levels of the filesystems and volume managers.)

 

[bsbeamer]

DriveDx on Mac side is great tool for monitoring drive health. Also builds in referential history to predict possible issues with both HDD and SDD. Works with external drives too, if you enable. This and Carbon Copy Cloner are two of my “must” tools.

 

[haralds]

I use SMARTReporter for menu monitoring and alarms and Disk Sensei for detail stuff.
CarbonCopy Cloner has been with me since its first beta!
[doublepost=1538329917][/doublepost]

DriveDx on Mac side is great tool for monitoring drive health. Also builds in referential history to predict possible issues with both HDD and SDD. Works with external drives too, if you enable. This and Carbon Copy Cloner are two of my “must” tools.

DriveDX does look good! But it just crashed on Mojave...

 

[killerovsky]

DriveDX does look good! But it just crashed on Mojave...

Please send crash report to support-at-binaryfruit.com

 

[haralds]

Please send crash report to support-at-binaryfruit.com

Happened only once and my logs were cleaned out since then.

 

[yifuhood]

I think there is a miss information about over-provisioning on Mac OS. if you have a Mac with T2 chip or a M1 Mac , over provision may not be necessary, but we don't know for sure.

if you have Mac that is earlier than 2015 or a SSD you put in yourself, for sure over provision works no difference as a PC. the SSD has its own controller chip and firmware. it is like Ireland is part of UK, but not part of great Britain, it has its own Government. with T2 chip and M1 Mac the SSD works under apple SSD controller. it is like China. I don't know apple algorithm.

so, you can set free space that is not allocated to any partition by Partition part of your SSD to Mac OS Extended, then First Aid this partition (it will trim this Partition), then reboot into ONLINE RECOVERY Option-Shift-Command-R. in disk Utility erase the Mac OS Extended partition. Done.

 

[h9826790]

I think there is a miss information about over-provisioning on Mac OS. if you have a Mac with T2 chip or a M1 Mac , over provision may not be necessary, but we don't know for sure.

if you have Mac that is earlier than 2015 or a SSD you put in yourself, for sure over provision works no difference as a PC. the SSD has its own controller chip and firmware. it is like Ireland is part of UK, but not part of great Britain, it has its own Government. with T2 chip and M1 Mac the SSD works under apple SSD controller. it is like China. I don't know apple algorithm.

so, you can set free space that is not allocated to any partition by Partition part of your SSD to Mac OS Extended, then First Aid this partition (it will trim this Partition), then reboot into ONLINE RECOVERY Option-Shift-Command-R. in disk Utility erase the Mac OS Extended partition. Done.

AFAIK, over provision is for GC to work more effectively. Otherwise, when the disk is too full, even with TRIM, no empty cell for the data to move around, GC still can't work effectively.

Regardless T2 or M1, SSD still the same SSD, still need TRIM and GC to maintain the write in performance, over provision still helps to make GC works better.

 

[yifuhood]

AFAIK, over provision is for GC to work more effectively. Otherwise, when the disk is too full, even with TRIM, no empty cell for the data to move around, GC still can't work effectively.

Regardless T2 or M1, SSD still the same SSD, still need TRIM and GC to maintain the write in performance, over provision still helps to make GC works better.

how would you know? what is apple ssd controller algorithm does to the ssd cells? it doesn't even has a Dram and controller for the ssd, but unifi memory and the CPU. and we know very little about APFS lower functioning. we don't know is we don't know.

 

[h9826790]

how would you know? what is apple ssd controller algorithm does to the ssd cells? it doesn't even has a Dram and controller for the ssd, but unifi memory and the CPU. and we know very little about APFS lower functioning. we don't know is we don't know.

What are you talking about?

That's just a NVMe, not unifi memory.

 

[yifuhood]

What are you talking about?

That's just a NVMe, not unifi memory.

It is not a simple NVMe ssd Especially with M1 Macs. A Typical NVMe ssd will have its own controller, higher end SSD will also comes with a ssd Cache memory chip(ssd dram). Which in the newer Macs those don’t appear. A Mac CPU - M1 is taking the tasks where other ssd need the ssd Controller for the tasks , in fact The CPU cache and system RAM are build in one, including the Storage CACHE (SSD DRAM). I think we should call it Embedded onboard storage memory, just like in a phone, in a tv, We don’t say a phone “ssd” storage memory is SATA Or NVME interface.

 

[h9826790]

It is not a simple NVMe ssd Especially with M1 Macs. A Typical NVMe ssd will have its own controller, higher end SSD will also comes with a ssd Cache memory chip(ssd dram). Which in the newer Macs those don’t appear. A Mac CPU - M1 is taking the tasks where other ssd need the ssd Controller for the tasks , in fact The CPU cache and system RAM are build in one, including the Storage CACHE (SSD DRAM). I think we should call it Embedded onboard storage memory, just like in a phone, in a tv, We don’t say a phone “ssd” storage memory is SATA Or NVME interface.

Ok, I got that part, but TRIM should be just TRIM, a function to tell GC which cell can be cleared. This logic should hold true regardless where the storage controller and cache is. Isn’t that a reasonable guess?

It’s known that even the M1 is using normal NVMe (the raw storage part). Which means the write in process should be no different than any other NVMe. Which means, they need TRIM (in fact, MacOS also call it TRIM on M1 Mac. Therefore, it's reasonable to expect it use the same logic to free up empty cells like normal TRIM does) to keep the write in performance. Their own controller may be more aggressive to use TRIM, or the other way around. That algorithm may be very different than other SSD, but so far, no evidence to show that TRIM itself on M1 has any difference than on other system.

And for GC to work properly，free space is required. And when the NVMe is nearly full, the only avail free space for GC to work effectively should be coming from overprovisioning. Even Apple hide many technical stuff, but this kind of basic stuff for non SLC solid state storage should be still true.

Of course, you can say this is BS. It's your opinion, I respect that. It's your freedom to think that way and express it accordingly. You are right that I don't know the details which Apple never disclose. But I think we can still have logical surmise base on what we know.

 

[Dayo]

when the NVMe is nearly full, the only avail free space for GC to work effectively should be coming from overprovisioning.

Is there a limitation that prevents RAM from being used? I suppose power loss during the process might be a problem.

 

[tsialex]

Is there a limitation that prevents RAM from being used? I suppose power loss during the process might be a problem.

TRIM is done by the SSD controller, not by macOS. When you delete a file and then the filesystem marks the sectors to be trimmed, the SSD controller receives this info and does all the reclamation of NAND sectors by itself and in the background.

 

[Dayo]

I suppose the query could have been more precisely framed as whether there is a limitation preventing the SSD Controller from using the RAM. Hardware can access RAM but not sure whether this is only before ExitBootServices.

EDIT:
Some DMA is possible: https://support.apple.com/en-gb/guide/security/seca4960c2b5/web. Seems technically possible but SSD Controllers might not be allowed (perhaps non Apple ones). Would be interesting to figure out but seems most likely to be an option for the controller.