Winmodem-like solid state storage
Posted Apr 11, 2009 19:00 UTC (Sat) by dwmw2
In reply to: Winmodem-like solid state storage
Parent article: Linux Storage and Filesystem Workshop, day 2
"But in the Winmodem route, the raw interface is a secret and specific to a small subset of devices, and the manufacturer doesn't write any Linux drivers. That's all pretty bad for Linux, isn't it?"
A WinModem is mostly just a sound card — we do actually know how to drive most of them. That's not the problem at all.
The problem with WinModems is that modem algorithms better than about v.32 are covered by patents. And while a lot of stuff in that situation has still found its way into software projects in the Free World, we still don't have a decent modem implementation. Someone needs to do a Free World fork of spandsp, perhaps? And/or pick up the late Tony Fisher's work on v.32bis and v.34.
For flash, the situation is different. While there are plenty of patents flying around, they mostly cover the ways in which you make a flash device pretend to be a block device — and the beauty of exposing flash directly to the operating system is that you don't need that gratuitous extra layer any more. You can have a file system which knows about flash, and is designed to operate directly, and optimally, on it.
The recent TRIM work goes some way to fixing the most obvious disadvantage of the extra layer, but the fact remains that you still have your real file system running on top of another pseudo-filesystem which is pretending to be a block device. And you can never attempt to debug or improve that lower layer.
The Linux kernel has two file systems for real flash already, and more are in the works. I'd very much like to see direct access to the flash being permitted by these devices. I'm confident that we can do better than anything they can do inside their little black box.
to post comments)