This is only an idea. I do not know if a software exists that would make this possible. Feedback is welcome.

 

1.) We create a share across multiple disks

2.) We use a software (don't know if it exist) to create a VHD, DMG, RAW disk image, etc which consists of as many files as disks are used by the share. Lets say 4 disks and 1TB image means 4x 250GB image parts.

3.) We copy the 4 parts to the 4 different disks

4.) We mount the image through the client.

 

 

Now we theoretically have a read and write speed 4 times faster than only with one disk.

 

2 hours ago, mgutt said:

Now we theoretically have a read and write speed 4 times faster than only with one disk.

What about parity? Are you proposing no parity disk(s) for this thought experiment?

 

If you are talking about parity protected array disks, then write speed is going to be way worse than any individual disk write, as each portion of the images being written is going to force the parity disk(s) to update multiple likely non-contiguous addresses.

12 hours ago, mgutt said:

Now we theoretically have a read and write speed 4 times faster than only with one disk

It may be possible to set this up, but even ignoring the limitations imposed by parity I do not see why you think this would improve performance.   You would still be only reading/writing to one of the vdisk files at any particular point in time - this type of arrangement does not give any sort of ‘striping’ which is the technique RAID uses to improve performance.

10 hours ago, jonathanm said:

If you are talking about parity protected array disks, then write speed is going to be way worse than any individual disk write

The writing speed depends on the size of the RAM (vm.dirty_ratio) and the size of the files that are written. But reads would be faster.

 

But as @itimpi pointed out correctly, I had a mistake in thinking. If you add files to a mounted image, those files will be written sequentially to the image. And as we have 4 huge image parts across 4 disks, it will only write to the first part as long its not filled and never touch the other involved disks. So to realize this idea it would be required to have much more parts across the disks, so even a small/medium file is written to multiple disks. This means the parts need a size of 1MB or so. ^^

Edited October 6, 20205 yr by mgutt