Solid-state storage devices and the block layer
Posted Oct 5, 2010 18:04 UTC (Tue) by jzbiciak
(✭ supporter ✭
In reply to: Solid-state storage devices and the block layer
Parent article: Solid-state storage devices and the block layer
I was thinking more in terms of treating flash specifically as less like an "I/O" device and more like a slow memory. I have no doubt that spinning rust will be around for awhile--a decade or more at least. It just seems like wrapping the flash behind a "disk drive" abstraction in hardware puts some artificial upper limits on how well it can perform. It's acceptable with spinning rust because the electronics are so much faster. When you go all solid-state, it just feels like a bottleneck.
Imagine what would happen if the immense creativity of the kernel crowd were unleashed on the problem of load balancing writes, erases and reads across a parallel array of raw flash modules?
Approaches such as UBI/UBIFS sound rather promising. I generally like the idea of owning the problem in kernel space, where it seems like we ought to be able do much more deliberate and proactive scheduling.
to post comments)