Re: SLS (was: native i/o and RCAC - performance)

Hello,

Am 14.05.2021 um 04:46 schrieb x y <xy6581@xxxxxxxxx>:

Physically, yes, but through an abstraction layer.

I know. :-) I've read Fortress Rochester & Co while ago.

The OS is unaware of the physical devices providing nonvolatile storage (magnetic disks and SSD's); the beauty of this approach is that underlying hardware changes (new storage technology) don't affect the OS and what we users (programmers) see.

That's how I understood the topic. And that's why I'm looking forward to non-volatile RAM technology market breakthrough: I'm curious to learn how IBM enhances its POWER machines with that. In theory, it'd be possible to eliminate DASD/SSDs completely. Everything is RAM, as Frank Soltis envisioned decades ago. The interesting bits will be: How to install/upgrade/IPL such a system?

On a side note, I've looked into NVDIMM implementation possibilities in Linux. As expected, it's… messy. Common OS's have the distinction of RAM and DASD built into their bones, thus options are limited.

One other consideration: after implementing a new device type, regression testing is way easier because you don't have to test the OS, just up to the interface.

This is true for other OS's, too. Have a look at Linux' device layering. It's not as rigid as IBM's approach, though.

And, honestly, the sheer size of today's SLIC with a lot of compatibility hacks isn't exactly what I'd qualify as "easier". :-) Even within SLIC, there can be quirks.

:wq! PoC