[Ham-Computers] RE: SATA

[Loren Moline WA7SKT]

Thanks Aaron and all..I understand now. Loren   WA7SKT Member: ARRL and Pacific Northwest VHF SocietyMember: Hearsat Satellite Monitoring GroupLocation: CN86cx                                                                                                                                                   > Date: Tue, 30 Dec 2008 20:54:04 -0800> From: aaron.hsu at nbcuni.com> To: Ham-Computers at mailman.qth.net> CC: > Subject: [Ham-Computers] RE: SATA> > Hi Loren,> > Yes, parallel data paths are quicker overall than a single serial data path, but parallel busses have inherent problems primarily related to timing. When you send data down a parallel bus, you need to make sure that every bit arrives at the exact same time - this is where the difficulties are comming in. As bus speeds get higher, the maximum trace/cable length shortens. Also, slight capacitance differences in each "wire" of the cable can delay the signal propagation of that one wire just enough to cause a timing problem (there's a specific word for this, but I don't remember what it is...something about latching, I believe). In the past, it was simpler to design a parallel bus to be faster because bus speeds were slower.> > But, with today's technology, things are getting faster and cable lengths and tolerances are getting shorter/tighter. It's already pretty difficult design parallel busses on motherboards because high bus speeds are bringing max trace lengths under 4 inches. If you look carefully at DDR/2/3 memory modules, you'll notice that some traces zig-zag or spiral for no apparent reason - but this is to make sure that each bit arrives at the exact same time at the motherboard.> > If I'm doing my math right, the max cable length for 0 wait states on a 100MHz bus (ATA-5, aka UDMA-100) is about 10 feet (assuming a 100% velocity factor). So let's say 6 feet for ATA-6/UDMA-133. Factor in cable VF and connector capacitance and the cable can now be no longer than 3 feet. Use a 50% factor to ensure timing can be accounted for and the max length is now 18 inches. Hey, guess what...that's the actual ATA-6 spec of 18" max!> > Today's technology is able to run at very high speeds, but the inherent timing problems of a parallel bus now becomes a major roadblock. Single-wire data-speeds of over 10GHz are possible, but trying to do this with a parallel bus is a nightmare. So, using a serial bus and appropriate timing protocols, you can run a single wire at 10GHz (for example) and get the full 10GHz bandwidth out of that cable (minus overhead, of course).> > Current SATA specs run at 1.5Gbps and 3Gbps (I believe they are baseband, so 1.5GHz and 3.0GHz). This would be a nightmare with a parallel bus, but pretty easy on a serial bus. This is the primary reason for the switch to serial architecture. But guess what, it's possible to "parallel" these serial datapaths to get "aggregate" bandwidth - I believe this is called "link aggregation". For example, the PCI-Express bus uses serial data lanes - but, you can combine up to 16 lines for higher bandwidth. PCI-Express x16 (16-lanes) is used by video cards and you'll find PCI-Express x4 (4-lanes) on server motherboards for high bandwidth controllers (hard drive and network). Why isn't this called a "parallel" bus? Because each serial "lane" is still transfering data independent of the others and a controller re-combines the aggregated data after-the-fact. In essence, it's a parallel bus...just doesn't work like a classic parallel bus.> > Anyways, I hope this helps clarify things. Serial buses are just easier to design at higher speeds. But, in essence, you'll still find "parallel"-like architectures using multi-link serial connections.> > If you haven't done so already, Wikipedia has a good article that covers some of this (though it doesn't really explain the technological hurdles of a parallel bus). On the bottom you'll find a chart comparing several of the most popular hard drive interface busses - the chart includes max speed, cable length, and number of devices supported.> > http://en.wikipedia.org/wiki/Serial_ATA> > 73,> > - Aaron, NN6O> > > > ________________________________> > From: ham-computers-bounces at mailman.qth.net [mailto:ham-computers-bounces at mailman.qth.net] On Behalf Of Loren Moline WA7SKT> Sent: Tuesday, December 30, 2008 6:04 PM> To: ham-computers at mailman.qth.net> Subject: [Ham-Computers] SATA> > > Hello,> > Can someone tell me why they have gone to Serial ATA drives?> > How can someone say that serial could be as fast as parallel data transfer.> > Just like USB external drives. No matter how fast USB is I can't believe it wouldn't be slower than parallel data transfer.> > > Loren WA7SKT> > > Member: ARRL and Pacific Northwest VHF Society> Member: Hearsat Satellite Monitoring Group> Location: CN86cx > > > > > > > > > > ______________________________________________________________> Ham-Computers mailing list> Home: http://mailman.qth.net/mailman/listinfo/ham-computers> Help: http://mailman.qth.net/mmfaq.html> Post: mailto:Ham-Computers at mailman.qth.net
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.qth.net/pipermail/ham-computers/attachments/20081230/37f8be3c/attachment.htm