Has mechanical drive technology stagnated?
- Thread starter Deep
- Start date
Not quite true, the more data per square cm, the faster the read/write rate at the same RPM.
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
Very true, you're ONLY talking about sequential speeds which are, have been and always will be by FAR the least important factor in a drives "speed".
A IDE drive from a decade ago, at 200gb and 90mb/s transfer rates had a 4kb random read speed of 0.3-0.4mb/s, a sata 3, 4gb hdd with 160mb/s max transfer rate has a 4kb random read speed of 0.4mb/s. 4kb random reads are the reason hdd's bog down when you unrar a large file, or try to play a game and unrar something at the same time, play a game and run an encode in the background, load windows, load applications do anything basically.
hdd's have gone from around 100mb/s to 160 mb/s or so, in like a decade or more, while random speeds haven't really increased in the slightest, while capacity has gone up 10-15x. SSD's have gone from 64gb drives that do 100mb read/writes with 4-8mb/s random speeds, to 512gb drives with 500mb/s read/writes and 30mb/s random reads and 100mb/s random writes.
HDD's are at the end of their capability in terms of raw performance and have been for 5+ years. SSD's are fast maxing out because while HDD ultimate performance is limited by rotational speed. Random performance is ENTIRELY dictated by the time taken to seek the next read/write, and beyond 7200rpm is both, noisier, higher failure rates, more expensive and uses more power. SSD's as they stand are limited by several things, still seek speeds to a degree as a 500mb's drive may do better in benchmarks yet give the same load times in most games and most applications, because the random read has been stuck around 30mb's for 2-3 years now.
Still ultimately down to latency, its magnitudes faster at maybe 0.1-0.3ms seek compared to 13-16ms for HDD's, ultimately you can't read the next file till you read, send and request another one and have another 0.1ms seek.
We will need new controllers and faster memory to go much beyond what we have now, sequentials is limited by both the interface, sata 3, which can be improved easily, chipset, relatively easy to improve but not much need yet, and the memory involved, which is approaching limits but newer memory should blow through them completely.
What we need for the next real leap in speed is a fundamental change in the way programs access data, storing things as larger files with batches of files it needs within, so loading a 2mb file at 500mb's rather than 100 4kb's files at 30mb's, better control of data streams and asking for more data at a time to increase the queue depth, etc.
Plus you should do regular backups anyway
While prudent, backups don't prevent failures, they just mitigate them. Having a drive fail is still annoying even if you have backups (cost of replacement, hassle restoring backups etc)
edit: as for the topic, I think the writing is definitely on the wall for mechanical drives. SSD should continue to increase in capacity at a given price point and although people are downloading a lot more content now, typical internet connections will probably reach a point at which people are less concerned about local storage. I know plenty of people for example don't worry about keeping everything they've downloaded stored, they just download it again if they want it (e.g. games on Steam). Streaming video seems a lot more popular these days too.
When you think about it, mechanical HD technology must be one of the oldest we have in terms of internal PC components, most of the other tech from 20+ years ago has been phased out.
Last edited:
That was a good read. I still confuses me why programs are not made to load the way you mentioned in the quoted paragraph above.
you are a tiny % of the market, 95% of peopel have a 80 -500gb drive in their machines and its more than enough....
Soldato
Mechanical technology is still improving in leaps and bounds.
To understand mechanical drive performance, you have to understand it's limitations. Many of them are material, including the weight and stiffness of the actuator arms (needed to move around the Read-write head), the substrate put on to the glass platters, and the read-write heads which do the reading and writing (funnily enough).
This isn't just figuring out a better way of doing something, this is coming up with new materials which are cheap and abundant enough to be mass produced which also do the job better.
Toshiba have just released a new 15K 300GB 2.5" SAS disk which offers 220 IOPs, rather than the previous best of 180 IOPs.
In the real world, that's roughly 20% more actions per second, or 20% less disks needed for the same amount of actions. For some companies, that might mean 4 racks of disks instead of 5, and all the energy (cost) savings that come from it.
There is still a lot of money in the market, and thus R&D will continue.
To understand mechanical drive performance, you have to understand it's limitations. Many of them are material, including the weight and stiffness of the actuator arms (needed to move around the Read-write head), the substrate put on to the glass platters, and the read-write heads which do the reading and writing (funnily enough).
This isn't just figuring out a better way of doing something, this is coming up with new materials which are cheap and abundant enough to be mass produced which also do the job better.
Toshiba have just released a new 15K 300GB 2.5" SAS disk which offers 220 IOPs, rather than the previous best of 180 IOPs.
In the real world, that's roughly 20% more actions per second, or 20% less disks needed for the same amount of actions. For some companies, that might mean 4 racks of disks instead of 5, and all the energy (cost) savings that come from it.
There is still a lot of money in the market, and thus R&D will continue.
Last edited:
Associate
- Joined
- 31 May 2012
- Posts
- 7
http://www.overclockers.co.uk/showproduct.php?prodid=HD-000-PL&groupid=701&catid=2104&subcat=
256GB for £125. Put 4 of them in RAID (or in a Windows 8 storage space so TRIM works), and you've got a TB for £500.
the technology has stagnated this year but there is still room for expansion. seagate are employing a new tech called heat assisted magnetic recording that looks to get 1TB/inch. i also recall there is some headroom left in the perpendicular magnetic recording method still used today.
article on xbit labs
http://www.xbitlabs.com/news/storag...ves_Densities_to_Double_by_2016_Analysts.html
cloud is a buzz word at the moment i think but until infrastructure in the country is built up. i dont see myself using it and i would much prefer to have my data myself. its odd to me that people who oppose having to be online all the time for games they have paid for (ubisofts methods for example) embrace cloud storage for other media they have bought.
article on xbit labs
http://www.xbitlabs.com/news/storag...ves_Densities_to_Double_by_2016_Analysts.html
cloud is a buzz word at the moment i think but until infrastructure in the country is built up. i dont see myself using it and i would much prefer to have my data myself. its odd to me that people who oppose having to be online all the time for games they have paid for (ubisofts methods for example) embrace cloud storage for other media they have bought.
Last edited:
Using the net for backing up is stupid, lazy and slow. I backup my important data (drive images,docs, music ect) on 2 hdds, so if 1 dies, its not a prob
Last edited:
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
There's little to no need, when you have incredibly effective caching, which you can have, you hide the latency, which is the only advantage a 10k rpm drive has over a slower, cheaper, quieter, lower power, more efficient, lower failure rate 5400-7200rpm drives.
I'd never in a million years buy an ssd for cache and pair it with a loud high power drive. You'd gain incredibly little performance going from a 20gb sdd + 5400rpm drive to a 20gb sdd + 10k rpm drive.
Hybrid drives are certainly interesting, ridiculously rare right now in terms of models, competition and capacity, meaning they can almost set any price they want, WAY cheaper than an ssd of similar capacity and WAY faster than a hdd of any kind.
WE need all the HDD makers(as few of them as there are) to get in the fight with hybrids and bring the price down. a 750gb hybrid is almost three times the price(before the price jump last year) of a non hybrid version, with what, £15's worth of nand flash in it you're paying £70-80 more. I'd have an ssd main drive and 1tb hybrid second drive in every single computer if they cost £20 more.
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
220 iops, in 300gb, and save one rack, but if a company needs a certain level of iops, you'd get 80-90k iops from a single ssd..... how many racks do you save at that point?
The newer ssd's iops was the biggest gain over any other performance metric vs last gen. 180 to 220iops, SSD's went from 20-30k iops to 80-90k iops this generation.
20% bump in iops is actually very very poor compared to the increase ssd's are making.
I'm not entirely sure why the industry is still so heavily invested in mechanical drives, there will likely be a lot of deals going on to subsidise the cost to keep sales going but ultimately at some point companies have got to switch because the difference is ridiculous.
@drunkenmaster
Sorry, My bad. I'm not suggesting that they fuse a 10K disk and a SSD together, just the idea of a 2.5" disk and a chunk of SSD storage into a 3.5" disk sized sled or box. The disk could be a 10K, 7.2K or 5.4K. My point was more relating to having a 3.5" disk unit that gave us somewhere in the region of a 1-2TB capacity with enough SSD capacity to make the performance of the unit far exceed a standard hdd. But as I'm writing this, there's also nothing stopping someone creating a 3.5" SSD that contains 1-2 TB capacity, which would definitely kill my hybrid idea, barring cost, ofc...
The problem I see with Seagate's Hybrid XT drive is that they chose to use SLC ram and only 8GB of it at that. Which makes it quite cost prohibitive at the moment until toshiba or others bring out their competing product. They also have the SLC cache set up as a read only cache, so I/O writes are really not making use of the hybrid design yet. (I've heard that they will be adding a firmware variant that may allow writes to use the cache though, not usre if it's available yet.) It might just be me, but I really don't think 8GB cache is enough...
As to mechanical HDD's in enterprise? I know at the moment, I have no choice but to use mostly mechanical HDD's. To get the required storage for our needs in SSD's we'd run out of disk bays and more importantly bankrupt the company. Using a san tiered cache to move the data up and down the disk types allows us to have sufficient IOPS whilst having a seamless large storage pool for data at a reasonable cost utilising all types of drives, SSD -> FC/SAS -> SATA. If the mechanical drives are gaining a percentage improvement of even 5%, I'll take it.
Sorry, My bad. I'm not suggesting that they fuse a 10K disk and a SSD together, just the idea of a 2.5" disk and a chunk of SSD storage into a 3.5" disk sized sled or box. The disk could be a 10K, 7.2K or 5.4K. My point was more relating to having a 3.5" disk unit that gave us somewhere in the region of a 1-2TB capacity with enough SSD capacity to make the performance of the unit far exceed a standard hdd. But as I'm writing this, there's also nothing stopping someone creating a 3.5" SSD that contains 1-2 TB capacity, which would definitely kill my hybrid idea, barring cost, ofc...
The problem I see with Seagate's Hybrid XT drive is that they chose to use SLC ram and only 8GB of it at that. Which makes it quite cost prohibitive at the moment until toshiba or others bring out their competing product. They also have the SLC cache set up as a read only cache, so I/O writes are really not making use of the hybrid design yet. (I've heard that they will be adding a firmware variant that may allow writes to use the cache though, not usre if it's available yet.) It might just be me, but I really don't think 8GB cache is enough...
As to mechanical HDD's in enterprise? I know at the moment, I have no choice but to use mostly mechanical HDD's. To get the required storage for our needs in SSD's we'd run out of disk bays and more importantly bankrupt the company. Using a san tiered cache to move the data up and down the disk types allows us to have sufficient IOPS whilst having a seamless large storage pool for data at a reasonable cost utilising all types of drives, SSD -> FC/SAS -> SATA. If the mechanical drives are gaining a percentage improvement of even 5%, I'll take it.