Future of the hard drive?

[Quitch]

When do people see the hard drive becoming obsolete?

The thought came to me when I looked up HDDs the other day. I didn't realise they were so cheap! 400GB drives available for less than 50p a gig! Yet it seems to me that while size is important, SCSI shows that it's not everything. Therefore I'm sure that eventually something else (flash memory?) will replace them as a narrowing of the price gap means that their advantages will outweigh their smaller size, until their increased popularity leads to a more rapid increase in size, until something replaces them...

Do people see HDDs becoming obsolete? Replacing them with something like flash memory makes sense as you lose the moving parts, remove issues such as fragmentation... but right now it costs too much, but that won't always be the case, as HDDs and RAM have shown.

What do you think and when do you think it will happen?

 

[Fruitfrenzy]

remove issues such as fragmentation...

Don't flash drives use the same file systems as HDDs? I would have thought fragmentation would be an issue independent of the physical implementation of the storage.

I have seen many references to all sorts of solid state holographic storage and stuff like that. I guess eventually one of these technologies will prove to be good enough to replace spinning disks but I wouldn't expect flash to. It is always going to be way too expensive to frabricate large chips in my (unqualified) opinion.

 

[Quitch]

Flash drives have, AFAIK, no moving parts. It's the moving parts that cause fragmentation to be an issue as data can only be read from where the head is located. The more the head needs to move the longer accessing data takes. If there's no movement (like in RAM) then it doesn't matter where the data is stored.

 

[Tokelil]

Flash drives have, AFAIK, no moving parts. It's the moving parts that cause fragmentation to be an issue as data can only be read from where the head is located. The more the head needs to move the longer accessing data takes. If there's no movement (like in RAM) then it doesn't matter where the data is stored.

You can only read from one place in (Flash)RAM as well. (Unless mulitiport) So you would still get penalty from fragmentation. (Though not as critical bacause of the much faster seek time)

 

[Fruitfrenzy]

Flash drives have, AFAIK, no moving parts. It's the moving parts that cause fragmentation to be an issue as data can only be read from where the head is located. The more the head needs to move the longer accessing data takes. If there's no movement (like in RAM) then it doesn't matter where the data is stored.

Ahh, I see your point. You still get the same level fragmentation but it doesn't slow things down to the same degree so you don't care.

 

[Cartoon Corpse]

what has better access? 1 300G drive or 2 150G drives? (all other things equal).

are 2 drives using 1 controller normally on pc's?

 

[digitalwanderer]

1 300G drive or 2 150G drives? (all other things equal).

I'd say the 2 150Gb drives. Ever since I added a second 200Gb HD I've noticed that you can really zip things up by keeping things on seperate drives. (80Gb IDE & 2x200Gb sATA right now in Bubbles, with a 30Gb USB in an external enclosure....I feel SOOOO spoiled right now! )


I keep azureus running on a seperate drive, my swap file on another, and try and keep windows on the IDE for the most part aside from that....it helps.

 

[Quitch]

what has better access? 1 300G drive or 2 150G drives? (all other things equal).

are 2 drives using 1 controller normally on pc's?

www.storagereview.com

Hard drive nerds

Assuming normal desktop use, I'd have to say 1 300GB drive would be superior.

Ahh, I see your point. You still get the same level fragmentation but it doesn't slow things down to the same degree so you don't care.

Well, as I understand it, since there's no moving parts it makes no difference to access speed whether the data is sequential or not, thus fragmentation becomes an irrelevance.

 

[akira888]

I'd say the 2 150Gb drives. Ever since I added a second 200Gb HD I've noticed that you can really zip things up by keeping things on seperate drives.

I keep azureus running on a seperate drive, my swap file on another, and try and keep windows on the IDE for the most part aside from that....it helps.

Have you have thought about creating a RAID 0 partition somewhere on your dual drives?

I have 4 300 GB drives w/NCQ and 16MB cache with one striped RAID 0 partition going across all drives and with my experience I will say that it is that it's easily worth with the effort. Just be sure to get NCQ on all drives though...

 

[Quitch]

Tell that to Storage Review and they'll tell you why NOT to do that

 

[Guden Oden]

Have you have thought about creating a RAID 0 partition somewhere on your dual drives?

Why would you want to do that? Raid0 does precious little to combat the biggest problem plaguing harddrives; seek times, and pretty much only doubles sequential data transfer rate, which has pretty much zero impact on application load times. On top of all that, you double the failure probability of your partition as well, spreading it across two drives like that. One drive dies = all data gone. Dumb idea, except for those into like, video editing or such.

Just be sure to get NCQ on all drives though...

NCQ tends to increase load times under light load circumstances (normal single-user access patterns). Again, not a terribly good idea for most people to use.

 

[Guden Oden]

Oh, and as for flash replacing harddrives... Never. Flash still only allows a guaranteed 100k erasure/write cycles. You'd wear out a flashdrive in months at most the way windows does spurious I/O.

Perhaps MRAM will be the vessel that'll replace harddrives, but it was a long time since that memory type made the news, unless something happens soon on that front chances are it'll be relegated to another of those techs that looked good on paper but wasn't feasible in reality.

 

[AlphaWolf]

Why would you want to do that? Raid0 does precious little to combat the biggest problem plaguing harddrives; seek times, and pretty much only doubles sequential data transfer rate, which has pretty much zero impact on application load times. On top of all that, you double the failure probability of your partition as well, spreading it across two drives like that. One drive dies = all data gone. Dumb idea, except for those into like, video editing or such.

The original comparison was 1 300gb vs 2 150gb, with the 1 300, 1 drive fails, all your data is gone. If you are worried about anything on your drives you really need to back up regularily or run a redundant array. If you are running 2 drives and aren't worried about mission critical data loss (I back up the small amount of important data on my drives regularily) a raid 0 array is pretty nice. I'd never go back to a non-raid setup and would consider adding redundancy in the future.

 

[Quitch]

If you're THAT worried about data then you wouldn't keep it all in a single location. What if your PC is nicked? What about a fire? Flooding?

Frankly, I'd advise you keep backups on disc, a pen or online somewhere.

Raid 0 really isn't going to give you much help. As stated, it helps only in sequential read times, which don't happen very often, and doubles the chance of a complete data loss.

 

[Gubbi]

For applications with lots of I/O transactions more disks (spindles) always mean higher performance. Depending on RAID type speedup will vary:

RAID 0: Speeds up both reads and writes, since writes and reads are spread out across the total number of disks.
RAID 1: Speeds up reads, write performance should be the same or lower. Reads are spread out across both mirror sets, writes has to write to both sets.
RAID 5: Speeds up reads and writes. Reads can be spread out across all spindles, similar to RAID 0. Writes has to write redundant data-block to disk.

For reference: I built the cheapest possible media serving box I could 2 years ago with a nForce 1 chipset motherboard (integrated video) + 1GHz Duron (left over) + 256 MB PC200 (left over) + 4 x 300GB PATA drives. Installed Linux on it and used software RAID 5 on the disks. Sustained read speed after two years of usage (ie. a bit of fragmentation) is ~100MB/s, write speed 55MB/s. Initially the figures were 120MB/s read and 60MB/s write.

Individually the drives are at 45MB/s read and 35MB/s write.

All measurements done with hdparm.

I'd love to rebuild it with beefier hardware, especially since I put a GigE NIC in it and can't get transferrate above 30MB/s (Duron+PCI33 showing it's limitation?)

Cheers
Gubbi

 

[Gubbi]

Oh, and as for flash replacing harddrives... Never. Flash still only allows a guaranteed 100k erasure/write cycles. You'd wear out a flashdrive in months at most the way windows does spurious I/O.

IMO, the only things in windows that sees heavy write action are the swap file(s) and the logs. Swap file won't be a problem, long before it becomes economical to replace HDs with flash, you'd have so much DRAM in your machine that swap partition/files make zero sense. That leaves logs. The OS would have to take care not to place often re-written data in the same locations (ie. use a new flash block each time logs are purged/rotated)

Look at it this way: If you have a 120GB flash disk with a write speed of 50MB/s (low, yes, but so is the capacity). Then, to write all of it 100,000 times would take more than 7 years of non-stop writing.

Perhaps MRAM will be the vessel that'll replace harddrives, but it was a long time since that memory type made the news, unless something happens soon on that front chances are it'll be relegated to another of those techs that looked good on paper but wasn't feasible in reality.

That's not going to happen, density seems to be a real problem. Largest capacity device is currently at 4Mbit, 3 decimal orders of magnitude behind flash (Samsung has a 4Gbit device).

Cheers
Gubbi

 

[Quitch]

For applications with lots of I/O transactions more disks (spindles) always mean higher performance. Depending on RAID type speedup will vary:

RAID 0: Speeds up both reads and writes, since writes and reads are spread out across the total number of disks.
RAID 1: Speeds up reads, write performance should be the same or lower. Reads are spread out across both mirror sets, writes has to write to both sets.
RAID 5: Speeds up reads and writes. Reads can be spread out across all spindles, similar to RAID 0. Writes has to write redundant data-block to disk.

For reference: I built the cheapest possible media serving box I could 2 years ago with a nForce 1 chipset motherboard (integrated video) + 1GHz Duron (left over) + 256 MB PC200 (left over) + 4 x 300GB PATA drives. Installed Linux on it and used software RAID 5 on the disks. Sustained read speed after two years of usage (ie. a bit of fragmentation) is ~100MB/s, write speed 55MB/s. Initially the figures were 120MB/s read and 60MB/s write.

Individually the drives are at 45MB/s read and 35MB/s write.

All measurements done with hdparm.

I'd love to rebuild it with beefier hardware, especially since I put a GigE NIC in it and can't get transferrate above 30MB/s (Duron+PCI33 showing it's limitation?)

Cheers
Gubbi

Actually RAID5 decreases write speed, as can RAID1. RAID1 only increases read if you have a good (read expensive) RAID controller. RAID0 increases read speed only in sequential reads.

I'd go state those views over at StorageReview and watch what happens

 

[Tahir2]

I will agree with Guden. Built a lot of systems with RAID 0 at customers request and it doesn't really do anything for your system except create more stress on your hard drives and increase the chances of data loss.
RAID 0 = not smart

 

[Gubbi]

Actually RAID5 decreases write speed, as can RAID1. RAID1 only increases read if you have a good (read expensive) RAID controller. RAID0 increases read speed only in sequential reads.

I'd go state those views over at StorageReview and watch what happens

RAID 5 does not decrease write speed. Every write has to hit two spindles, yes. But you very seldom uses RAID 5 on two drives. With three or more drives in your set, write speed increases. Also: See my numbers backing it up.

RAID 1 increases read speed in software RAID under Linux. You have twice the spindles to read from, so must be a pretty crappy controller to actually degrade performance.

RAID 0 should also increase non-sequential reads, since you have twice the spindles and hence twice the seeks/s. Only exception is pathological cases where all seeks hits the same disk.

Cheers
Gubbi

 

[Quitch]

Those numbers don't go into enough depth to have meaning. You should duke it out at their site I mentioned, they know much more about RAID than anyone here I suspect, and I respect their views on these matters.