Lo! All. Every had an idea for a new piece of hardware? Been daydreaming of something that would send your peformance through the roof? Planning the next computer revolution? If you feel like sharing any ideas relating to new or improved kit then chuck them down in here. Bit about your new toy, how/why it would make things better, anything atall to get the idea across. Share! --------------------------------- --------------------------------- Jaz_knos
Lo! Might as well start the ball rolling. A picture says a thousand words so: Basically a dual (or multi) head hard disk drive, allowing shorter access times, faster read/write speeds, symultanious reading and writing, generally uber performance data storage and retrieval. Coupled with some updated NCQ technology and a large ammount of buffer space (less that £20 for a 1Gb of flash ram so why not make use of the speed for OS startup??) should give something that is currently not offered by RAID arrays. Enjoy! Jaz_knos
It isn't marketable, practical and doesn't offer a performance advantage. - It isn't a standard form factor when 3.5" disk internals are used unless it gets bumped up to 5.25". Lets pretend for a minute that they just stretch a standard disk out a bit: sure this isn't a problem for modders, which make up a part of a fraction of a percent of the market. Average home users wouldn't know what to do with it and most enterprise OEMs and customers would never use a product that is anything but completely standardised. Lets pretend its a 5.25" drive now. This makes it technically viable for home users however enterprise users are still left in the cold. You can fit 2.5x the number of 3.5" devices in the space that 5.25" devices needs. Even if the performance improvement you claim is gained with a device like this, it can't make up this deficit. - Comments have been made about this improving access time. While this is true, there are 2 points to make here. Firstly, in a home system, access time makes very little difference to performance because there really isnt very much seeking. Instead of going into it here, its already been discussed to a much further extent than I would be bothered to here (be warned, its pretty heavy stuff). Secondly, average access times at best would improve by around 15%, hardly cutting them in half. The average rotational latency on a 7200rpm drive is around 4ms. Your proposed design would cut this rotational latency in half, saving 2ms. Considering that the norm for average access times on a 7200rpm drive is 13ms, 2ms isn't much fo a saving at all. Compare this to 7.5-8ms for 10k or 5.5-6ms for 15k drives - 11ms isn't anything special (from a device likely to cost as much as a 10k unit anyway). - from a practicality standpoint, the more moving parts there are the more chance of something breaking. The added electonics and extra head would also increase heat, further reducing reliability. - Theoretically, sure, you can double throughput. Though, it would be pointless having both actuator stacks reading from the same track (just reading the same data twice, so sequential throughput would be no different to that of a single drive. Given that single-user disk access is highly localised, this starts to put a dent in the potential advantages of multi-head drives in a desktop environment. In a server environment, forgetting the form-factor problem for the moment, RAID has become very well established indeed and modern disk arrays can deliver uttely massive amounts of bandwidth and IOs. The simple scaleability of hard drives means that if you need more performance, you get more drives. If high throughput is the primary aim from storage, RAID0 would be more effective. - The electonics would be far more complex, and drive the cost up considerably (manufacturing cost may not be much higher, but R&D costs would be massive). There aren't many people willing to absorb this cost, in general, people don't care about the performance of a drive (or don't care enough to spend significantly more). Just to re-iterate the above, in short: more spindles > more actuators Last of all, don't you think that if it was such an excellent idea this would have been done? Guess what, it has (mid 80s iirc). The company flopped. The future is tiered storage. A level of fast solid-state storage in between the mechanical disk and volatile cache can solve just about every gripe about modern hard drives that consumers could have. It would offer the speed of solid state memory where it needs it and the mass capacity of magnetic platters when it doesn't. EDIT: great, after I've typed this out I've found it already done for me here
Redundant Array of Inexpensive Platters is a pretty funky concept, especially as you could configure them as Just a Bunch of Platters (JBOP)
Well it'd go well with the master / slave non-PC terms we have already The the new Hitachi 1TB drive that features five platters would make for a nice RAID-5 array, 750GB of redundant (and fast) storage. Again the added complexity and cost unfortunately negates the advantages
The question remains, would you RAIP your master, RAIP your slave, or just go ahead and RAIP both of them??
RAIP sounds like a pretty dumb idea to me - if the drive electronics fritz, then you've got a dead array. The point of the redundancy in RAID 5 is that you can recover all the data after the failure of ANY single device (platter, drive electronics, cable, controller, PSU etc.). The chance of a single platter in a drive failing without the rest of the unit also failing is pretty slim.
I think the idea of it is to provide a RAID0-like sequential transfers. The main drawback, however, is that very few applications make large 100% sequential transfers. Seeing that a single actuator stack can only access a single cylinder at a time this is the only area that this platter striping would be of any benefit.
RAIP seems like a very efficient way to half your storage, whilst providing no protection from anything other than read error (even then you don't know which platter has the correct data on it) Dissapointed the multi-headed-HDD has been pwned but once again science and maths win out
