Using Sata Ports As A Network Bridge

[Scytheon3]

When building consumer or commodity clusters, one of the main bottlenecks is the network. This is because the only network that people can reasonably afford is GigE as things like infiniband are way to expensive per node. SATA III now has a transfer rate of 6 GB/s so it is 6 times as fast as GigE so why is it not possible to use this as a form of communication between two or more computers.

Something very similar to this is the USB network bridges you can buy (example) but they are much slower. When searching on Google the only thing I could find was this but it is only for SATA I and it only seemed to be a theory with no actual working prototypes.

If I tried to do this how would I accomplish it? Could I simply plug a Male to Male SATA cable into the two computers and read and write directly to the SATA port? Can you actually directly read and write to a SATA port? On the USB network bridges there is some kind of circuit in the middle, what does this actually do and how would it be implemented for SATA?

[Infiltrator]

That's an interesting theory, would be nice to see a switch capable of transferring data at the speed of Sata 6.

[abferm]

I would think it would be more an issue of writing custom drivers. Maybe some custom hardware as well.

[buffy]

Just go 10GbE its not that expensive, and your cables can be longer.

[buffy]

Or for really really cheap ass got for USB3.0 to USB 3.0 5Gig

http://www.geartaker.com/product/usb-3-0-pc-to-pc-device-data-transfer-link-sync-cable-blue-150cm

$3.99

[digip]

Only problem with the USB bus is for every other USB device plugged in, the speed drops between each device to share bandwidth on the bus lanes. Not sure how the specs work on USB 3.0 though, but I imagine thats one of the reasons they went with a faster protocol, so multiple connected devices wouldn't see such a noticeable difference in speed transfers. You can take a USB hub and mesh a room of computers together, but not very practical. USB 3.0 might work better, but I'm not even sure if PC to PC works with more than two machines. Would be neat to try it out though, since parts would be cheap. There is also the limitation on cable length with USB though. I don't remember off the top of my head but I think its like 16 feet?

[atblakely]

The idea is interesting, but for SATA as a network, you'd defiantly have an interesting time with it. My guess is that you simply can't plug two computers together (if you did a null modem approach, maybe.) but then the issue is the SATA controller is expecting a SATA device, not another controller. You'd probably have to design some kind of device which emulates a SATA device on both ports then does the required IO multiplexing. If you'd like to actually make an implantation I'd start out by reading the SATA protocol spec. Who knows, there might be something in there that would make this easier.

[Rampage]

i know you are gonna flame me for this answer, but why not considering thunderbolt instead of SATA?

i think due to its nature it might be easier to implement network over thunderbolt.

i know as of now thunderbolt is spread only across Mac systems, but motherboards for PCs with thunderbolt are starting to appear from brands like asus.

and i honestly hope this technology spreads because it's really great :)

[barry99705]

i know you are gonna flame me for this answer, but why not considering thunderbolt instead of SATA?

i think due to its nature it might be easier to implement network over thunderbolt.

i know as of now thunderbolt is spread only across Mac systems, but motherboards for PCs with thunderbolt are starting to appear from brands like asus.

and i honestly hope this technology spreads because it's really great :)

Drivers are the issue. AFAIK there are no thunderbolt/light peak linux/unix drivers. Since most clustering services use these two operating systems, it's not going to work.