SSD 4K speed measures when formatting in 64K

[majidok]

Hello,

So all SSD specs are in 4K , what happens if I format my partitions in 64K clusters? Im sure there would be a speed increase even though I probably lose more space ? and those speeds can be theoretically multiplied by 16?

 

[USAFRet]

Hello,

So all SSD specs are in 4K , what happens if I format my partitions in 64K clusters? Im sure there would be a speed increase even though I probably lose more space ? and those speeds can be theoretically multiplied by 16?

No, your SSD speed will not magically increase by 16x.
It might even decrease.

You could try it, and show us the before and after results.

 

[majidok]

Hello,

So all SSD specs are in 4K , what happens if I format my partitions in 64K clusters? Im sure there would be a speed increase even though I probably lose more space ? and those speeds can be theoretically multiplied by 16?

No, your SSD speed will not magically increase by 16x.
It might even decrease.

You could try it, and show us the before and after results.

I can guess it wouldn't increase by 16x , but we should see improvement or not? and since you are saying it might even decrease you have seen it first hand or somewhere you can link ? And why are iops measurements in 4K ? I'm sure average file size is above that

 

[USAFRet]

Hello,

So all SSD specs are in 4K , what happens if I format my partitions in 64K clusters? Im sure there would be a speed increase even though I probably lose more space ? and those speeds can be theoretically multiplied by 16?

No, your SSD speed will not magically increase by 16x.
It might even decrease.

You could try it, and show us the before and after results.

I can guess it wouldn't increase by 16x , but we should see improvement or not? and since you are saying it might even decrease you have seen it first hand or somewhere you can link ? And why are iops measurements in 4K ? I'm sure average file size is above that

No, I have not personally seen this, nor seen any tests of this.
However...your supposition of 'average file size is above that' is almost certainly wrong.
Text files, fonts, etc, etc. All tiny.

And that 4k is not file size, but rather block size.

As said...test this and show us.
Select a large selection of mixed data. Maybe 20GB os so.
Large files, small files, tiny files. How long does that take to transfer?

Then, reformat that SSD into 64k blocks.
Using that exact same sample data...how long does it take to transfer?

Normal cleanroom testing rules apply.


In building the current OS's, and how things are generally formatted...and in building current drives...there are a whole lot of big brained people heavily invested in making things 'fast'.
Trust me...they've looked into this.


Oh, and your premise seems to be aimed solely at SSD's. Why would this also not apply to spinning drives?

 

[rkzhao]

The standard in the industry for benchmarks is 4K transfer on random operations and 128K on sequential operations. What that means is each individual write or read command carries 4K/128K worth of data. This is just a benchmark standard so there's some common measurement unit, with 4K being the smallest aligned transfer length and 128K being a large enough transfer to make firmware overhead irrelevant. Your system can send commands to the drive in any multiple of 512 bytes (LBAs).

Performance specs report sequential performance in MB/s and random performance in IOPs. This is in part because MB/s on random 4K operations is not as impressive so it looks better to use a command count. As your transfer lengths get higher on randoms, your IOPs will go down since each command is transferring more data but the MB/s will generally go up.

Essentially for most drives, the 4k random speed represent the highest possible IOPs that the drive can achieve and the sequential performance will be the highest MB/s.

 

[majidok]

Hello,

So all SSD specs are in 4K , what happens if I format my partitions in 64K clusters? Im sure there would be a speed increase even though I probably lose more space ? and those speeds can be theoretically multiplied by 16?

No, your SSD speed will not magically increase by 16x.
It might even decrease.

You could try it, and show us the before and after results.

I can guess it wouldn't increase by 16x , but we should see improvement or not? and since you are saying it might even decrease you have seen it first hand or somewhere you can link ? And why are iops measurements in 4K ? I'm sure average file size is above that

No, I have not personally seen this, nor seen any tests of this.
However...your supposition of 'average file size is above that' is almost certainly wrong.
Text files, fonts, etc, etc. All tiny.

And that 4k is not file size, but rather block size.

As said...test this and show us.
Select a large selection of mixed data. Maybe 20GB os so.
Large files, small files, tiny files. How long does that take to transfer?

Then, reformat that SSD into 64k blocks.
Using that exact same sample data...how long does it take to transfer?

Normal cleanroom testing rules apply.


In building the current OS's, and how things are generally formatted...and in building current drives...there are a whole lot of big brained people heavily invested in making things 'fast'.
Trust me...they've looked into this.


Oh, and your premise seems to be aimed solely at SSD's. Why would this also not apply to spinning drives?

Because I already did that on mechanical drives more than 10 years ago and yes there was performance improvement , you might even find my thread here . also I'm here for answers to that not to provide answers , but yes i wanted to do this and I will and will post the results , ofc i wanted to know if anyone has done this before , also just because "a whole lot of big brained people" as you call them are doing something doesn't mean new ideas cant come to peoples minds

 

[USAFRet]

I (and probably many others) will be very interested in seeing your test results.

 

[majidok]

I (and probably many others) will be very interested in seeing your test results.

Very nice , sure I will try it , also I know that 4K is block size , but well files go into blocks . So so far no result from anyone who tried this it seems

 

[rkzhao]

Well technically 512 is the standard block size. For HDDs, 4K block size was added to improve density and performance. Most 4K drives are still emulated down to 512 byte LBAs. This is where the idea of 4K alignment comes in where commands should start at the beginning of a 4K physical block for optimal performance on 4K devices. Some of that has carried over to SSDs for some reason but I don't really know what benefit there is to 4K SSDs.

In anycase, in theory, there really shouldn't be much of a reason for performance to go down if you go above 4K, assuming commands are still 4K aligned. Going back to what I mentioned earlier, going to a higher transfer length will decrease IOPs but generally increase MB/s. Given that what people actually care about with performance is MB/s, it wouldn't be surprising for you to see some performance increase. What you're playing with is how much individual command overhead contributes to the total amount of data transfered.