Messages in this thread | | | Date | Sun, 13 Feb 2000 16:48:40 -0800 (PST) | From | Jeremy Fitzhardinge <> | Subject | Re: Scheduled Transfer Protocol on Linux |
| |
On 13-Feb-00 Brandon S. Allbery KF8NH wrote: > In message <200002132112.NAA09873@work.bitmover.com>, Larry McVoy writes: > +----- >| : I wasn't so much interested in that level as the one where SCSI is itself >| : a >| : networking protocol. Current SCSI drives are effectively IDE drives with >| >> ( ... ) >| >| Am I missing knowledge of some advance in SCSI? > +--->8 > > Again, I'm not talking about protocols. I'm addressing the argument that an > STP-capable drive is somehow fundamentally different from a SCSI-capable > one, since that's the argument that keeps being raised against this idea. > I'm talking about the controller board, not the details of communication.
It is not very different. The point is that SCSI over STP over Gigabit ether is a better physical connection than SCSI over fat cables or SCSI over fibrechannel.
The underlying point is that: - drive manufacturers want to sell more drives - they get practially zero margin on current IDE drives - you can only attach 4 drives to a typical PC with IDE, and you need to open the case to do it; IDE represents approx 100% of the market though - SCSI has pretty expensive fiddly cabling, and is only marginally more flexible than IDE; hardly anyone has SCSI controllers, but you can charge more for SCSI drives - fibrechannel is really expensive (the connectors alone can be ~$100), it's pretty flexible; it has ~0% of the market, but you can charge a fair amount for drives - 100baseT is cheap, very flexible, ubiquitous, easy to use; the only problem is that nobody makes drives which talk to it, but it makes a lot more sense than SCSI, IDE and FC (at least for single drives with ~10MB/s transfer rates) - ditto Gigabit ether, except that at the moment it's more expensive and less common; there's a much higher chance of it becoming common than FC however.
STP is simply an enabler which makes it worth considering ethernet, since ethernet would otherwise have too much latency and pipeline bubbles to make it effective.
J
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
| |