0

I searched many reviews and descriptions about SSD specifications but couldn't find convincing information.

Write durability of the memory cells TBW seems to be the most important, but how to review its effect on usage when most normal users like me use the large capacity of the HDD/SSD to store data for a long time without change, so most of the SSD space will be written no more than few times.

The real problem is the System drive(s) which also contain(s) the swap area which will be written thousands of times for swap data and lots of times for considerable system and application files. Considering this I don't see a big difference between SSD brands and designs.

If we want 1 TB SSD and choices for TBW are 300/4500/600/750 means the more affected memory cells could be rewritten between 300-750 times both are less than expected rewrite for 1 year work.

But SSD were working for much more than that, so I expect that a SSD controller could detect the failing cells and reallocate the addresses to not affected banks, is that true? If that is the case then the total durability for the SSD will be very long, until the controller reallocation database became very large and stop functioning, of course the SSD speed will decline as the reallocation DB become larger, if such technique is really exist "it does not exist in RAM".

r2d3
  • 3,298
  • 1
  • 8
  • 24
Hamed
  • 31
  • 7
  • 2
    SSDs use wear leveling which means frequently written portions (such as swap) won't be written to the same place, they'll be spread out evenly across all the cells. – At0micMutex Dec 12 '21 at 04:17
  • @At0mic I expected that, so why making such importance of TBW, and how this work? does it move the swap only to free spaces, means the controller will read the file system, or it chose space and swap the data between swap and the original data whatsoever. – Hamed Dec 12 '21 at 04:21
  • 2
    The controller doesnt read the filesystem, it works below that level, presumably keeping track of cells that have not been written. There is also this concept of overprovisioning - ie part of the SSD is not visible to the OS/filesystem and is used to spread writes. – davidgo Dec 12 '21 at 04:26
  • What is true of the relationship between os and ssd is that deletes dont actually delete stuff they just mark the blocks as free to the ssd (google TRIM ssd) – davidgo Dec 12 '21 at 04:29
  • @davidgo Thanks for the comment, then the real task for user like me is to understand each controller approach, because the controller can't keep record of write on each sector, so they are using blocks could be 4kB or 1MB, smaller means more process and much slower, but much convenient for some work uses lots of temporary small files like programming – Hamed Dec 13 '21 at 07:14
  • 1
    @Hamed I wonder if you are overthinking it. (In 2021) Its generally more useful to look at the type of memory you are getting then the controller - the type of memory and amount of over-provisioning define the longevity of the drives. I dont think any drives would use 1MB blocks. For durability stay away from QLC based disks (ideally you want SLC but these are $$$ so TLC is probably a good trade off. If you purchase a bigger disk then you need it will wear level over a larger surface area and last longer as well. That said durability is not the problem its often made out to be) – davidgo Dec 13 '21 at 09:01
  • 1
    You say "If we want 1 TB SSD and choices for TBW are 300/4500/600/750 means the more affected memory cells could be rewritten between 300-750 times both are less than expected rewrite for 1 year work." This is doubtful. Enterprises run heavy database loads for years without issues. A typical TLC will last way more then 5 years as a swap drive on a typical server. On a desktop its a non-issue. I don't think you are taking wear leveling into account. – davidgo Dec 13 '21 at 09:13
  • @davidgo "This is doubtful. Enterprises run heavy database loads for years without issues" yes I have desktops running OS on Sandisk Q64G for years, but recently the SSD became slower no failure or bad sectors so I start researching – Hamed Dec 13 '21 at 14:05
  • 1
    @Hamed - Yup your ssd needs replacing, but comparing a very early controller with tiny memory and to a modern one with 15+ times the space + more overprovisioning is unfair. Ive personally run postgres databases for SaaS providers for years so maybe I have some idea what Im talking about. – davidgo Dec 13 '21 at 18:57
  • As for best practices - databases on SSD - Amazon Web Services database offerings typically run on SSD - https://docs.aws.amazon.com/AmazonRDS/latest/UserGuide/CHAP_Storage.html evidences this Azure similarly runs SSD on their databasrs. – davidgo Dec 13 '21 at 19:11
  • 1
    Also read https://superuser.com/questions/51724/should-i-keep-my-swap-file-on-an-ssd-drive - **but read past the outdated accepted answer from 2009** and look at the heavily upvoted subsequent answers. – davidgo Dec 13 '21 at 19:19

1 Answers1

1

1. Comments regarding your posting

Write durability of the memory cells TBW seems to be the most important, but how to review its effect on usage when most normal users like me use the large capacity of the HDD/SSD to store data for a long time without change, so most of the SSD space will be written no more than few times.

Your use profil is not hitting the boundaries of the SSDs. If you want to "review" you have to test the SSDs by reading and writing continuously until they break.

The real problem is the System drive(s) which also contain(s) the swap area which will be written thousands of times for swap data and lots of times for considerable system and application files. Considering this I don't see a big difference between SSD brands and designs.

This is a false conclusion. If you accept your conlusion asking the question becomes useless. The use of a SSD as a drive where the operating system and swap data resides makes a failure more likely.

If we want 1 TB SSD and choices for TBW are 300/4500/600/750 means the more affected memory cells could be rewritten between 300-750 times both are less than expected rewrite for 1 year work.

How do you get to the implicit conclusion above that you are writing 1 TB per day?

But SSD were working for much more than that, so I expect that a SSD controller could detect the failing cells and reallocate the addresses to not affected banks, is that true?

The controller does not want to detect failing cells. He does everything to distribute wear evenly by internally reassigning cells.

If that is the case then the total durability for the SSD will be very long, until the controller reallocation database became very large and stop functioning, of course the SSD speed will decline as the reallocation DB become larger

There is no need for a database. There is just one table needed which assigns LBA sector numbers to flash cells and location therein. Furthermore a table containing a write counter is necessary for flash cells or group of flash cells. Both table do not grow in size when being realized as an array.

, if such technique is really exist "it does not exist in RAM".

That does not matter.

2. Recommendations

Your SSD selection is determined by your preferences which are unknown for us.

Here are some issues you can think about:

  1. Can I afford the loss of SSD data? I only use SSD as a system drive. I can recover my private system the long way using installation media and a collection of software (no business continuity issue) or the fast way by using a complete system backup which I have but would have to be updated.

=> I can afford the loss of SSD data.

  1. Price per Terabyte written

The following lines assume that most private computer users rarely make a backup - like me.

  1. Readibility of the state of wear? I compared SMART parameters of SSDs from Adata, Crucial and Western Digital. Although they share some common attributes, each manufacturer uses parameters that the other do not use.

  2. Your use strategy: a) Not writing the full TBW storage figures creating additional breakdown security b) Exceeding the TBW hoping to exploit a safety margin from the manufacturer

  3. Test have shown that overusing a SSD result in different types of breakdowns. Some SSDs remain still readible, other fail completely. Would you pay higher prices for products that only convert to read-only?

  4. Assuming that you are faced with a breakdown you can research for which flash controllers the data recovery scene has already complete time-saving rescue solutions.

r2d3
  • 3,298
  • 1
  • 8
  • 24
  • why should I test the machine to a limit they will never reach? such test is used for the type of the Nand when they develop. – Hamed Dec 13 '21 at 07:19
  • Under assumption the addresses are fixed like RAM the swap area will be written 10s of thousands of times or more in a year, but as they commented above addresses are not fixed. – Hamed Dec 13 '21 at 07:26
  • Yes that evenly distribution approach is what we discuss earlier – Hamed Dec 13 '21 at 07:30