Slashdot Mirror
Western Digital Working On a 20,000 RPM Drive
MrKaos writes "Western Digital seems to be preparing for the onslaught of solid-state drives set to impact its market by developing a 20,000 rpm hard drive. Similar to the VelociRaptor line of drives, the new drives are speculated to be offering lower capacity as a tradeoff for faster seek and write times." This report out of Taipei is the only word on the rumored WD 20K drive. It's said to be a 2.5" drive in a 3.5" enclosure, for efficiency of cooling — the arrangement the Register enjoyed poking fun at when the 10K drive was upgraded last month.
Apparently http://en.wikipedia.org/wiki/Solid-state_drive answers my questions:
Price - as of mid-2008, flash memory prices are still considerably more costly per gigabyte than are comparable conventional hard drives: around USD 3.50 per GB[10] compared to typically less than USD 0.40 for mechanical drives.[11]
Capacity - although currently far lower than that of conventional hard drives, SSD capacity is predicted to increase rapidly, with experimental drives of up to 1 TB in test.[12][13]
Higher vulnerability to certain types of effects, including abrupt power loss (especially DRAM based SSDs), magnetic fields and electric/static charges, in comparison to normal HDDs (which store the data inside a Faraday cage).
Limited write cycles - flash-memory cells will often wear out after 10,000-100,000 write cycles[citation needed], while high endurance cells may have an endurance of 1-5 million write cycles (many log files, file allocation tables, and other commonly used parts of the file system exceed this over the lifetime of a computer.[14] Special file systems or firmware designs can mitigate this problem by spreading writes over the entire device (so-called wear levelling), rather than rewriting files in place.[15] Today's drives can last up to 20 years with average usage.[dubious - discuss] An example for the lifetime of SSD is explained in detail in this wiki.[dubious - discuss] SSDs based on DRAM, however, do not suffer from this problem.
Slower write speeds - as erase blocks on flash-based SSDs generally are quite large, they are far slower than conventional disks for random writes and therefore vulnerable to write fragmentation,[16] and in some cases for sequential writes.[6] SSDs based on DRAM do not suffer from this problem.
Lower storage density - hard disks can store more data per unit volume than DRAM or flash SSDs, except for very low capacity/small devices.
Higher power consumption at idle or under low workloads laptop battery runtimes decrease when using an SSD over a 7200 RPM 2.5" laptop hard drive,[17] flash drives also take more power per gigabyte.
RAM based SSD require more power than hard disks, both operating and when turned off.[18]
Well, that would add to the number of components that could fail, and require a high speed bus between the two controllers, as well as a shared cache and all the headaches that would bring with it (think SMP caches being ping-ponged). Then you've got to sync your interface to the system bus as well as the new internal buses. On the other hand, you can just crank the knob up to 11 and go 20K RPMs on known, tried and true, technology.
If I mod you up, it doesn't necessarily mean I agree with what you've said, sorry.
Yeah, it's the last time I'm going to buy one of those things... until the next sale... or if I need to pick one up at Wal*Mart at 3AM. I've been saying that for three years now. Somehow my drives die just in time for the newest WD's to be on sale.
If I mod you up, it doesn't necessarily mean I agree with what you've said, sorry.
The higher speed drives aren't so much for their sequential transfer rates by themselves, but their random seek rates. They are trying to get high I/O per second rates (IOPS), which is what a lot of servers need to be at their peak.
No matter how dense the data is, random access speeds are dominated by how long it takes to move the head to the data and how long it takes to wait for the data to rotate under the head. A smaller platter will mean it takes less time to move the head on average, but the only way to get the data under the head faster is to increase rotational speed.
A 7200 rpm drive has an average 8.3ms rotational latency; a 15k rpm drive is 4ms, and a 20k rpm drive is 3ms. In other words, this speed increase could enable the drive to do 10% more random I/Os per second.
dom
From what I remember reading, they actually split the laser into several beams that could be read by a single assembly - that way if you were reading one long continuous file, it could effectively increase your read speed since it was reading 3 or more chunks at once.
Random access wouldn't benefit from this as much; for maximum speed you would need to be reading something that spanned the length of all 3 tracks (or more) at once. Writes would be even more interesting in this scenario.
Karnal
The smaller diameter / mass will tend to reduce bad effects from conservation of angular momentum.
Those are my principles, and if you don't like them... well, I have others.
Google might find a use for SSD (using caching and stuff), but your average business running an Oracle database to keep track of their business information wants read-write (not read) performance, which is what these things can do.
that bit about power consumption was partially recanted by Toms.
apparently they didnt do their homework well enough.
color me surprised.
not to mention the article basically says that current drives have almost no power saving features and performance was on par to resulting in slightly more consumption, whereas platter drives have had decades to develop power saving features.
i expect this 'result' to be completely wrong in the next couple product cycles. the intel 'mass market' drives already advertise significant power saving potential and are the first consumer component offered.
turn up the jukebox and tell me a lie
Connor actually did this right around the time 3.5" drives started.
http://en.wikipedia.org/wiki/Conner_Peripherals#Performance_issues_and_the_.22Chinook.22_dual-actuator_drive
It could read from either set of heads, but I believe could only write from one set. Writes can be posted in a write-behind buffer, so this didn't impact performance.
http://lkml.org/lkml/2005/8/20/95
Who will be famous as the first consumer to die in a hard-drive failure?
Table-ized A.I.
You don't get +5 Funny for "getting your science right".
It's been done before, iirc, but they tend to be more expensive, and the multiple heads run the risk of creating unintended harmonics. Most of the time it would be cheaper and faster to use two drives with one set of heads, than one drive with two sets of heads.
The problem is that going from 15k drives to 20k drives means going from roughly 150 IOPS to 200 IOPS.
That's it.
SSDs are getting 20,000 IOPS, and some specialized SSD and RAM systems like FusionIO and Violin memory are getting 120,000 to 1,000,000 IOPS (these are not typos)
Sure, SSDs are expensive but spinning disks are beginning to look more and more like an evolutionary dead end, at least for IOPS.