An idea regarding HDDs

[ViperV990]

I've had this idea for a long time. I've never heard anything about it tho... I'm sure it is technically possible, but I'm not sure if it is economically feasible. But then of course, a Radeon 9700 pro is no where near to be economic...

But anyway, here's my idea: A drive with 2 independent heads, one of them accessing the one side of a platter while the other the other side. Basically, a RAID 0 array in a single drive.

If done correctly, this should yield close to twice the performence. Downside is of course the resulted additional cost and possibly lower reliability, but hell, for performence's sake, someone outta make a drive like this...

So, is this idea feasible? Any other pro and contra that I haven't touched? Replies are, of course, highly welcome...

 

[Demon-Xanth]

It's actually been done. But DOS didn't support it so it never caught on.

 

[ViperV990]

Why didn't DOS support it? Shouldn't the drive appear totally transparently as a single drive?

Well anyway, gasoline direct injection was introduced back then with the MB 300SL in the 50s, but it wasn't until a few years ago that Mitsubishi brought GDI on the market (in Japan and Europe anyway)... and now quite a few European and Japanese auto makers have caught on with GDI.

 

[ViperV990]

Also, you have any links to that old drive? I'd like to read 'em myself...

 

[Demon-Xanth]

I don't have a link (these were rare, and before the internet). But the problem with GDI in the US is excess NOx emmisions when they make them run lean to squeeze in some more efficency.

 

[ViperV990]

The problem, AFAIK, can be solved by lowering the sulfur concentration in the gasoline, or so I heard...

But before that thread gets too off-topic, the point I was making should be clear: I believe right now there's a market for a dual head HDD. I believe a lot of people wouldn't mind paying 25% extra for an 80gig drive that's almost twice as fast than the normal ones.

 

[TerryMathews]

But it wouldn't truly be twice as fast, as the second head would be limited by the rotation of the drive. I.E. on a 2 drive RAID, each drive has it's own platter and platter spin motor. The drive can spin the platter forward or backwards, depending on where the FAT table says the information is located.

Your system could increase throughput, but the drive would have to be 100% defragmented all the time to get all the performance, and it would be twice as susceptible to fragmentation. Which would show itself in slow seek times.

What I would like to see is the IDE bus supporting disconnects during seeks, so that the other drive can talk a la SCSI. Maybe Serial ATA? Just think, you could actually do a 4 drive RAID0 on two controllers with no slow down.

 

[MajinVegeta]

Another thought about HDD performance why not include DIMM / SO DIMM slots on it for an expandable buffer? forget this 2-8 meg tiny buffer garbage how bout a 2-8 Gig Buffer tat you decide what files get cached.

 

[ViperV990]

But it wouldn't truly be twice as fast, as the second head would be limited by the rotation of the drive. I.E. on a 2 drive RAID, each drive has it's own platter and platter spin motor. The drive can spin the platter forward or backwards, depending on where the FAT table says the information is located.

No way, a running HDD's rotation speed is constant (or at least very close to that).

Another thought about HDD performance why not include DIMM / SO DIMM slots on it for an expandable buffer? forget this 2-8 meg tiny buffer garbage how bout a 2-8 Gig Buffer tat you decide what files get cached.

A SO-DIMM slot occupies too much height. Plus I'm skeptical about how much additional performence it will bring.

 

[stebesplace]

Perhaps this may be close to what you are looking for

http://www.explosivelabs.com/articles/sata/pg2.shtml

Serial ATA, which is much faster than parallel.

-Steve

 

[Peter]

Well the interface is faster. The drives are not.

Double-head-assembly drives have been done, Seagate's very first Barracuda series were like that. However since the read channel circuitry and head assembly are the expensive bits in a HDD, these drives cost twice as much as normal ones. The next idea, and a better one I must say, was RAID. Which made these two-headed drives disappear soon after.

 

[Peter]

On IDE, that'd be useless anyway, since IDE (with VERY few exceptions) does not support out-of-order command execution. All accesses are fully in-sequence, meaning that even if you have two head assemblies (or two fully separate drives on the same IDE channel), you wouldn't be able to process two commands at the same time.

btw, Serial ATA cannot even use two drives on one cable anymore.

 

[ViperV990]

Originally posted by: Peter
On IDE, that'd be useless anyway, since IDE (with VERY few exceptions) does not support out-of-order command execution. All accesses are fully in-sequence, meaning that even if you have two head assemblies (or two fully separate drives on the same IDE channel), you wouldn't be able to process two commands at the same time.

But what if the drive has some kind of controller that makes it totally transparent to the outside world (i.e. the IDE controller)?

But anyway, you've convinced me that RAID 0 is a better, cuz less expensive, alternative.

 

[ViperV990]

Found something about the Seagate drive:

LINK

 

[Shalmanese]

This question pops up ever 6 months or so. Basically the reason are:

1. Its too expensive
2. Theres no demand
3. Most importantly, heat from the drive cause the platter to expand non-uniformly, screwing up the read heads.

 

[Peter]

But what if the drive has some kind of controller that makes it totally transparent to the outside world (i.e. the IDE controller)?

That still wouldn't allow this unit to process more than one request at any given time, so there's no point in having two head assemblies.

SCSI is different, having tagged command queuing and disconnect/reselect strategies on the interface level. The combination of both allows drives to take further requests while working on the previously received ones, and even allows them to complete them in any suitable order, thus minimizing head movement and rotational latency. (Plus it frees up interface bandwidth so that multiple drives on the same cable can actually share the interface effectively.)

But even with SCSI, two normal drives in striping RAID would give the same improvement as would such a dualheaded special drive.