[Info-vax] VMS and the (lack of the) TRIM facility.

[JF Mezei]

On 15-06-20 11:35, Stephen Hoffman wrote:

> As mentioned elsewhere, HP already supports their own SSDs on OpenVMS, 
> and those have their own connections as well as what HP calls a "wear 
> gage", so presumably some sort of TRIM-like support is present though 


Nop. SSDs do behave as normal disks to the OS. The relocate pages
(groups of blocks) to different physical locations when a single block
is rewritten. But the OS still sees virtual block numbers as if they
were physical.

 So consider when you issue 5 IOs to rewrite 5 consecutive blocks.  The
page that contains these blocks will be moved and rewritten 5 times. So
5 pages get "wear".

If the page which contains those 5 blocks had been marked as free with
TRIM, it would be zapped once (the wear) and each block written once,
leaving a few unused blocks ready to be written to (depending on page size).

This requires the OS tell the disk when a block is free. And when the
SSDs sees all blocks in a page are free, that page can be zapped and
used without needing to preserve blocks into another page when this page
is being written to.