USB 3.0 Mid-2009 - 625MB/s transfer rate!

[Candlelight]

http://en.wikipedia.org/wiki/Serial_ATA#eSATA_in_comparison_to_other_external_buses

Compared to USB 2.0's poxy 60MB/s.

That will certainly speed things up a bit. Wonder what problems we'll face though.

 

[Nick Ryder]

Well what would suck is if they only made USB 3.0 backward compatible with 2.0 and not 1.0. Not everyone has 100% access to a newer usb powered computer. Although I suppose if they make USB 3.0 Cardbus cards and USB 3.0 upgrade pci cards that would be good. 60MB is still fast, but yeah 625 would be almost faster than a hard drive, definately put the nails in the mechanical hard drive market. Once they can get us solid state hard drives in larger sizes - 100-200 gigs in prices that are close to what you can get a traditonal drive for... that'd be a glorious day. Far less heat output, more durable computers, far easier to upgrade and/or secure data.

 

[Rii]

^ 625MB/sec is much faster than any hard drive on the market, mechanical or solid state.

 

[Candlelight]

Yeah, eSata is only rated at 300MB/s.

 

[Rii]

The fastest mechanical HDD on the market is the Seagate's 15k RPM Cheetah which hits 135MB/sec read speed on the outer tracks. There appears to be at least a dozen different SSD manufacturers each claiming that they produce the fastest SSD around, but I haven't seen any hard numbers above 200MB/sec there either. Write speeds are lower than read for both mechanical and solid state drives.

Looking at performance only, SATA is overkill for most consumers, the additional bandwidth offered by "SATA II" is of no real use to anyone. They're fine standards, certainly superior to PATA, but raw bandwidth isn't high up on the list of reasons why. Lower production costs, easier cable/airflow management, increased resistance to electromagnetic interference and subsequently reduced error-checking requirements, longer cables, less susceptible to physical damage from improper insertion/removal, one device per channel, traditional SCSI features like staggered spin-up, etc. are the real benefits SATA brings to the table.

 

[Marc]

^ 625MB/sec is much faster than any hard drive on the market, mechanical or solid state.

I can't remember the details but supposedly USB while offering 400MB/sec strangles the drive performance but it's just not suited to high volume file transfer.

Prior to the advent of E-Sata, going Firewire offered better performance (and while they finally getting USB 3.0 to 625MB/sec we've already got Firewire @ 800 but then I guess there's something to be said for a closed standard

 

[Rii]

^ Yeah, USB 2.0 often doesn't get anywhere near it's theoretical 480Mbit/sec transfer rate in practice, so I've no trouble believing the same is/will be true of USB 3.0. It should be fast enough that it won't be the limiting factor for any of the tasks your average consumer is likely to call upon it for though.

 

[Nick Ryder]

But I imagine using a USB external hard drive would be a much quicker experience. I mean hey for us old farts with hard drives under 80 gigs could practically use a half TB drive as a 'main drive' and barely notice. Now that's sweet

 

[Jadzia]

I like the idea of optical communications -- it is the way of the future -- but I don't like them piggybacking it onto usb. It just doesn't sound right to me. It should've been a separate new connector imo, purely optical, with no electronic component.

And leave usb alone -- windows struggles enough with 2.0 without tacking another data line on there.

 

[BCI]

Prior to the advent of E-Sata, going Firewire offered better performance (and while they finally getting USB 3.0 to 625MB/sec we've already got Firewire @ 800 but then I guess there's something to be said for a closed standard

I think some units are mixed up here. There is Megabytes per second, and there is Megabits per second. Which is which in the examples?

 

[Candlelight]

Prior to the advent of E-Sata, going Firewire offered better performance (and while they finally getting USB 3.0 to 625MB/sec we've already got Firewire @ 800 but then I guess there's something to be said for a closed standard

I think some units are mixed up here. There is Megabytes per second, and there is Megabits per second. Which is which in the examples?

From the chart in the link I posted:

Bus = MBit/s - MByte/s
Firewire400 = 393 – 49.13
USB 2.0 = 480 – 60
Firewire800 = 786 – 98.25
eSATA = 3000 – 300
USB 3.0 = 5000 – 625

 

[Brent]

As the speed increases we'll be moving over even more to solid state flash type devices like USB sticks and such. One move closer to the Isolinear world of Star Trek.

 

[farmkid]

As the speed increases we'll be moving over even more to solid state flash type devices like USB sticks and such. One move closer to the Isolinear world of Star Trek.

Even with USB 2.0, the interface isn't what limits the speed, the speed of the memory is the limiting factor. Even if your USB flash drive was USB 3.0-capable right now, it wouldn't be any faster in a USB 3.0 port. USB 3.0 will be useful for external hard drives since the USB interface is the bottleneck there, rather than the speed of reading to and writing to the platter. USB 3.0 will allow one to set up a striped raid array in an external enclosure to get faster data transfer rates than what is possible with a single drive.

 

[Jadzia]

USB 3.0 will allow one to set up a striped raid array in an external enclosure to get faster data transfer rates than what is possible with a single drive.

Raid-like technology could be built into hard disks, through use of multiple read-write heads arranged around the disk. Conventional heads sweep across 1/4 - 1/3 of the disk, so 3 or 4 heads would be the optimal setup. It would be easier to configure than conventional raid arrays, whilst being backward compatible with conventional drive controllers by simply not using the secondary heads.

 

[farmkid]

USB 3.0 will allow one to set up a striped raid array in an external enclosure to get faster data transfer rates than what is possible with a single drive.

Raid-like technology could be built into hard disks, through use of multiple read-write heads arranged around the disk. Conventional heads sweep across 1/4 - 1/3 of the disk, so 3 or 4 heads would be the optimal setup. It would be easier to configure than conventional raid arrays, whilst being backward compatible with conventional drive controllers by simply not using the secondary heads.

It's interesting you would mention that. That is something that occurred to me a couple of years ago. I mentioned it to a friend of mine and he told me about some HD company that experimented with that concept a few years ago. As I remember, they used it briefly, but dropped the idea. I don't remember why; it could have been that the drives were more prone to failure, or perhaps the performance gains weren't justified by the increase in complexity. Whatever, the reason, the idea didn't last long. Personally, I think it's a good idea that has promise.

 

[Candlelight]

It probably reduced the lifespan of the drive considerably, and they opted for a more stable system.

 

[Jadzia]

What is it that normally fails on a hard disk?

In my proposed system, if one of the heads fails (or it's controller electronics fails), it has these backup heads. Although it would operate less efficiently with less heads, all of them would have to fail before the drive was useless and the data was lost.

 

[LaxScrutiny]

USB 3.0 will allow one to set up a striped raid array in an external enclosure to get faster data transfer rates than what is possible with a single drive.

Raid-like technology could be built into hard disks, through use of multiple read-write heads arranged around the disk. Conventional heads sweep across 1/4 - 1/3 of the disk, so 3 or 4 heads would be the optimal setup. It would be easier to configure than conventional raid arrays, whilst being backward compatible with conventional drive controllers by simply not using the secondary heads.

It's interesting you would mention that. That is something that occurred to me a couple of years ago. I mentioned it to a friend of mine and he told me about some HD company that experimented with that concept a few years ago. As I remember, they used it briefly, but dropped the idea. I don't remember why; it could have been that the drives were more prone to failure, or perhaps the performance gains weren't justified by the increase in complexity. Whatever, the reason, the idea didn't last long. Personally, I think it's a good idea that has promise.

Hard drives have multiple platters, already with multiple heads. I realize you are talking about multiple heads per platter, but the simplest method would be utilizing all the platters simultaneously.

Especially since if you were doing a built in RAID, all the platters already spin at the same time whether they are being read/wrote or not. So data would would read/write to each platter in one spin cycle. There's no real cost or extra wear and tear.

The main thing to add in would be a controller to turn it into a RAID array, which adds to cost and also size and heat, just looking at an existing PCI RAID controller and imagining it all built into a hard drive.

 

[Jadzia]

I'm probably wrong on this, but it was my understanding that the heads move together for each platter. The idea being that as each byte is written, the 8 bits are written across 2 sides of 4 separate platters simultaneously.

Similar to ram. Each chip has a 1 bit width, and 8 chips are written to simultaneously across the bus, in order to write 1 byte.

I'm probably wrong on this, and I've never studied computer science, but this is what sounds like the most natural implementation to me.

 

[LaxScrutiny]

No, at present only one head reads/writes at a time. I can only guess why, simultaneous head use would need a more sophisticated controller on board to split data and create a large virtual drive out of up to 10 physical disks. This would be a RAID controller in addition to the conventional HD logic board you see on the bottom of the drive.

Right now there is basicly just a switch that sends data through one arm to one head to one platter, then the next, then the next as you go through sectors on a disk. To use each head simultaneously, you would have to split the data and then send each bit in sequence to 10 heads. Like I say, it's not impossible, it's already built into RAID controller cards, or done in software (slower). But a good RAID controller card is a couple hundred dollars and physically it's as big as the existing logic board on a drive. A drive that did this would have to bigger and hotter than what we have now.

I think it would be a workable idea, but there may be more issues I haven't thought of. Consider the WD Raptor, people will pay a a premium for a drive that's faster and hotter and noiser but faster. But mostly the demand is for more and more storage space.