Quote:
Originally Posted by Stinger
I graduated as a Electrical/Computer Engineer last year, so I fully understand the challenges of memory coherence in todays parallel computing world. But come on, with the wide proliferation of multi-core processors, the techniques in this massive field have been refined quite a bit. I actually worked with with some high-performance distributing computing clusters ( Sharcnet) while I was at university, which was very cool and interesting. |
At the risk of getting wildly off-topic and technical, this isn't the problem. The problem is that you would have two devices accessing one filesystem. Common operating systems use read/write caches on those filesystems, so what's on the disk doesn't necessarily match what's in memory.
Imagine two scenarios: device A has made changes to the FS in memory, but hasn't written them to the device when device B makes an update, then device A writes those cached versions that are now stale; and device A has a cached read of the FS that it's referring to to base changes on when device B writes to that area of disk behind its back. These would happen all the time, and would cause things to break in spectacular and interesting ways.
These aren't insoluble problems (NFS does this kind of thing in a very distant way, after all), but they're also not easy, and things like performance would suffer as a result (at the simplest level because you'd need to introduce locking, and methods to keep cache coherence.) That besides, Windows and MacOS aren't going to support this in a hurry, and fat32 never will, so that'd be another problem. You'd have to write your own drivers for them, and it'll no longer be anything close to 'plug'n'play'. (Not to mention that people like me would probably start to complain, as I bet you wouldn't release the specs so I could make it run in Linux
