From what I can gather TRIM is performed when an OS that supports TRIM deletes a file on an SSD that supports TRIM marking the block irrelevant so it won't be reallocated/rewritten upon next GC. (Controller doesn't know what data is valid/invalid, only discerns between empty/used blocks and once written will preserve any data except if OS tells it to overwrite a block or - via TRIM - tells it this block no longer contains valid data after deletion.)
That means TRIM has to be done when a file is deleted and can't be done afterwards.
If you want to TRIM a drive that's used with a non-TRIM OS the only way is to write data to all free space, then delete it on the TRIM OS.
Did I get that right?

InjecTioN wrote on 2020-01-16, 06:44:

🤦 you are correct. I have just checked and it was in fact turned off... How could I miss that!? Must have been the time.

I'll post some new screenshots tonight.

All good. I learned the hard way when I benched my own WinME machine with an SSD. It seemed super fast because it still had the ultra low response time, but it was maxing out at like 3.5MB/s. Since files were so small then, the response time really made up for it. ...I basically saturate the ATA133 bus with DMA enabled however.

This might be interesting: ATA TRIM command (low-level programming)

> Did I get that right?

Not really. The OS always knows which blocks are in use and which are free, all it needs to do is use TRIM on all the free blocks and it results in the same. But for win98 on a modern SSD...who cares?

The Serpent Rider wrote on 2020-01-15, 17:13:

Rudolf had utility to run TRIM in pure DOS, but it has some limitations.

link?

also I've been looking into this further - if you can take the drive out of the PC and/or dual boot into linux of some quasi-modern variant you can use fstrim to force a discard on all unused blocks of a mounted filesystem, or you can use blkdiscard to discard all blocks on a device (essentially a full refresh, with full data loss)

https://archive.org/search.php?query=cr ... ph+R.+Loew"

Considering that CF cards as PC storage has long been a option, and SATA to IDE is far from a new technology. I feel like if there was a common situation where using an SSD in Win(X without trim lead to catastrophic results, we'd have many stories about it now? The problem seems largely academic, and if a serious concern, likely limited to 'heavy usage' machines, daily drivers that do a lot of work. I think most of the contemporary use cases for Win9X on an SSD doesn't really fit that scenario.

120GB SSD is, frankly, the largest drive that Windows 9X can even deal without without relying on multiple partitions. In relative terms, even 120GB is a 'near bottomless hard drive' for most retro PC users. It'll be hard to really stress one of these drives running witout TRIM. I'm sure someone is going to show me a Win9X machine with 500 games installed, but can we agree that that's likely an edge case?

> I feel like if there was a common situation where using an SSD in Win(X without trim lead to catastrophic results, we'd have many stories about it now?

Win98 went out of widespread use around 2002. SSDs did not really hit mainstream until 2007(?). So no, we would not hear about it. And only early SSDs suffered from write exhaustion, new drives can withstand more than you are ever likely to throw at it, especially with a low-write-load OS like Win98.

my interest in this is nothing to do with longevity, it's all about avoiding slow copy-erase-write cycles on well used cards/drives

InjecTioN wrote on 2020-01-16, 00:58:

I have done some benchmarks with some tools that are available on https://www.philscomputerlab.com/hdd-benchmarks.html […]
Show full quote

InjecTioN wrote on 2020-01-15, 17:10:

We'll just have to test it then. 😀 I mean, it's definitely worth it if it in fact works. If it's just a placebo effect we're talking about, it's quite useless. I haven't read into TRIM enough to know exactly how it works though, so I'll start with that to get some actual results.

I have done some benchmarks with some tools that are available on https://www.philscomputerlab.com/hdd-benchmarks.html

Test results are in!

Disk full, no trim:
full.jpg

Disk empty, pre trim:
empty pre trim.jpg

Disk empty, post trim:
empty post trim.jpg

Facts!

Fujoshi-hime wrote on 2020-01-16, 12:41:

InjecTioN wrote on 2020-01-16, 06:44:

🤦 you are correct. I have just checked and it was in fact turned off... How could I miss that!? Must have been the time.

I'll post some new screenshots tonight.

All good. I learned the hard way when I benched my own WinME machine with an SSD. It seemed super fast because it still had the ultra low response time, but it was maxing out at like 3.5MB/s. Since files were so small then, the response time really made up for it. ...I basically saturate the ATA133 bus with DMA enabled however.

So, DMA obviously helped quite a bit.

With DMA, filesystem pretty full, no "trim":

With DMA, filesystem emptied using Explorer, no "trim":

With DMA, filesystem emptied using Explorer, with "trim":

In conclusion: Speed difference between a full drive and an empty drive is measurable but not noticeable in real world experience. When using the ForceTrim.exe application, there is no real difference in speed. It doesn't do anything useful for anything older than Windows 7.

These work great and only cost about $32 shipped from China (if you're willing to wait). They're 32 GB in size, IDE, and will plug directly into an IDE port in your motherboard.

I've also built and sold now over 100 systems, with WIndows 98, using the 44-pin variant (for thin clients) of this drive in 16GB size. They're a very reliable drive - I buy them used, and so far have yet to receive a single one DOA or have a single report from a buyer of a failed drive.

maxtherabbit wrote on 2020-01-16, 18:07:

my interest in this is nothing to do with longevity, it's all about avoiding slow copy-erase-write cycles on well used cards/drives

Right, but again, that concern is purely academic. No one here is seemingly able to cite real world practical examples of consequences. One could even set up a modern system and simply have it run writes and deletes in a simulated work load and measure performance changes over time.

More over, such hardware in the application has performance to spare. You'll be either running it natively over ATA133, or SATA1, so a drive that could normally hit 500MB/s will only have 133-150MB/s of actual bandwidth available in a 9X system. That SSD could lose have it's sequential read or write speed but that wouldn't be the bottleneck.

This seems to be largely the fear of an idea where the practical consequences have yet to even be demonstrated and measured. Everything is theoretical.

yeah ok except when you're talking about a small <16GB CF card that has shit for linear transfer rate to begin with it's a whole different story

maxtherabbit wrote on 2020-01-16, 19:08:

yeah ok except when you're talking about a small <16GB CF card that has shit for linear transfer rate to begin with it's a whole different story

Is it? All the same, at this juncture, no one seems to know what the story is. There's minimal documentation on it. And CF cards have lived in all kinds of industrial tasks, used on cameras for decades, writing piles of data. So are there any stories of serious problems?

I'm just saying that without documentation of catastrophic consequences, this is all just guessing, and maybe it's really not such a big deal? And if it is a big deal, then someone will hopefully document it when it happens. So I say plow right along as is, don't worry, and if something goes wrong... Write a nice forum post about it so other people Googling the same thing can find it.

I really don't understand why anybody, other than those obsessed with being "period correct," would want to use anything other than an SSD in a retro build. They're superior in just about every way imaginable. And all the issues raised here are moot IMO as it is pretty easy to just back up your drive once in a while and clone over to a new SSD if your currently installed one ever fails or slows down to molasses. They boot faster, they're silent (compared to the super noisy mechanical drives of the era), programs launch faster, they draw less electricity, they generate less heat; I could go on and on.

Fujoshi-hime wrote on 2020-01-16, 19:33:

maxtherabbit wrote on 2020-01-16, 19:08:

yeah ok except when you're talking about a small <16GB CF card that has shit for linear transfer rate to begin with it's a whole different story

Is it? All the same, at this juncture, no one seems to know what the story is. There's minimal documentation on it. And CF cards have lived in all kinds of industrial tasks, used on cameras for decades, writing piles of data. So are there any stories of serious problems?

I'm just saying that without documentation of catastrophic consequences, this is all just guessing, and maybe it's really not such a big deal? And if it is a big deal, then someone will hopefully document it when it happens. So I say plow right along as is, don't worry, and if something goes wrong... Write a nice forum post about it so other people Googling the same thing can find it.

I can tell you with certainty that my 4GB CF card that has been in use on a variety of retro tasks is now showing notably reduced write speeds, while the read speeds are unchanged from when it was brand new

It that a "serious problem" or "catastrophic consequence" certainly not, but diminished performance pisses me off

bmwsvsu wrote on 2020-01-16, 19:47:

I really don't understand why anybody, other than those obsessed with being "period correct," would want to use anything other than an SSD in a retro build. They're superior in just about every way imaginable. And all the issues raised here are moot IMO as it is pretty easy to just back up your drive once in a while and clone over to a new SSD if your currently installed one ever fails or slows down to molasses. They boot faster, they're silent (compared to the super noisy mechanical drives of the era), programs launch faster, they draw less electricity, they generate less heat; I could go on and on.

Not that I particularly care one way or the other, some people don't consider the experience complete without using a mechanical drive. They may see solid state as being too fast, unnaturally so, or they miss the noise that comes with those clunky boys. There's no one right or wrong way to enjoy your retro systems. What works great for me, for example, may be abhorrent and insane to you. Personally I use SD cards in an IDE adapter in both my 486 laptop and my Pentium Pro system, they're even cheaper and more than fast enough (and still have much quicker access times) and I'm eventually going to figure out how big of an SSD my PIII needs and clone it over to one as well (I haven't got my full software load on there figured out yet). Mechanical drives make me feel trapped for some reason.

Sedrosken wrote on 2020-01-16, 21:56:

Not that I particularly care one way or the other, some people don't consider the experience complete without using a mechanical drive. They may see solid state as being too fast, unnaturally so, or they miss the noise that comes with those clunky boys.

BRB, inventing 'Hard Drive Sound Generator' and putting it on Etsy. Like those 'Engine Sounds' for electric cars. 😮

^^ here, sample this one.

maxtherabbit wrote on 2020-01-16, 19:50:

I can tell you with certainty that my 4GB CF card that has been in use on a variety of retro tasks is now showing notably reduced write speeds, while the read speeds are unchanged from when it was brand new

It that a "serious problem" or "catastrophic consequence" certainly not, but diminished performance pisses me off

I could see that happening on 4gb drive, especially if you have more than 3gb of files on it. Wouldnt take an unreasonable amount of writes to have the wear leveling paint itself into a corner. Might be time to consider getting a newer 16GB Cf card or even better, getting an IDE to SD adapter with a 64gb SD card. Either route will set you back $25 but it might be $25 well spent. You will be CPU or ATA limited for a long time after making that purchase.

If you don’t want to spend the money, consider backing up the drive, doing a secure erase, and restoring.

https://www.pcworld.com/article/2088341/how-t … erformance.html

Yes, simply deleting files and repartitioning and formatting your drive won’t do the trick as it will with a hard drive. Those operations take place at levels above where true garbage collection occurs in an SSD. In fact, due to the total absence of utilities that force complete garbage collection, there’s only one way to return a heavily used SSD to pristine, like-new condition immediately—the ATA secure-erase command.