What is Digg?Sponsored by Sony Pictures
Do you believe the 2012 Mayan Prophecy? view!
whowillsurvive2012.com - The Mayan Calendar predicts the end of time: 2012. See the trailer for 2012, opening November 13.
35 Comments
- badenglishihave, on 04/13/2009, -0/+41And I want $10,000. Oh wait, same thing.
- lewystud, on 04/13/2009, -1/+32just want a 1TB SSD
- tortfeasor2009, on 04/13/2009, -2/+22I wish I had 24 of these:
http://www.youtube.com/watch?v=96dWOEa4Djs - Snappysnap, on 04/13/2009, -1/+16I can't keep up with SSD developments.
- aegis17, on 04/13/2009, -0/+11This article reads like it's straight out of 2007. Many of the "issues" it presents have been well-known for years, and many of the solutions it claims have been solved by Intel or OCZ in their latest SSD Vertex series. If there's truly anything actually innovative and new in this article, forgive me, my attention wanders when I read something I felt I've read a dozen times before in months and years past.
Still, dugg for SSD's kicking ass in the computer hardware market. - arunforce, on 04/13/2009, -0/+7Redundant reply is redundant.
- frequentFlyer, on 04/13/2009, -0/+6Now if they could only fix "endurance, reliability and data retention problems" in humans, we'd be all set!
- byronm, on 04/13/2009, -1/+6This is one technology that i'm anxiously waiting for. Anyone have any experience with these on windows media center - especially for a live tv/time shifted tv? does work like that cause an early grave or degradation on SSDs?
- FTLJohnson, on 04/13/2009, -0/+5Actually, I think you just made it a raid 5. Crap, that means I made it RAID 6.. or possibly raid 0+1...
and yes... it was bad... continuing it... worse. - Smono, on 04/13/2009, -0/+5RAID 1 reply is redundant.
...wow, that was bad. - Hello1024, on 04/13/2009, -0/+3I really don't see how endurance or performance is a problem in SSD's. It's all down to writing a good translation layer between the OS (which expects a fully reliable addressable block device) and the hardware (which is unreliable with some particular undesirable characteristics like slow erase or large blocks).
Now, a translation layer is FIRMWARE, and does a pretty simple task - effectively all you have to do is write one function which takes an OS address for a read/write, and figures out where in flash to put it, taking into account known bad blocks, pre-erased blocks, partially filled blocks etc.:
long mapAddresstoFlash(long LBAAddress) {...};
Now since the performance\reliability of the SSD is entirely dependent on this function, it's pretty important to get it right - so why not simply make a framework and a test harness and get a load of university students to each have a go at writing the best function for a known benchmark - call it a competition and give the winner (based on a benchmark score including lifetime and performance) a decent prize, and I guarantee within a few weeks work someone will have prototyped an algorithm that solves the problem better then any of the in-house programmers. Now all that needs to be done is convert that algorithm into something that can run on the SSD manufacturers hardware, and you're all ready to go. - ryan00davis, on 04/14/2009, -0/+3as someone who works as an engineer for hitachi gst in the enterprise division, i am scared of SSDs
- soulpunisher, on 04/13/2009, -1/+4be careful when buying these there are two kinds out MLC and SLC the SLC is much faster and intel puts out one of the fastest ones out...only problem is that it is only 64GB max right now :(
SLC intel
http://www.newegg.com/Product/Product.aspx?Item=N8 ... - xxmjumpman23xx, on 04/13/2009, -0/+2True, hopefully the cost of SSDs will decrease soon, new tech is always expensive as hell.
- pureliquidhw, on 04/13/2009, -1/+3thank god i'm not the only one
- Leviathan433, on 04/13/2009, -0/+2SNNNNNNAAAAAKKKKKKEEEEEEEEE
- JQP123, on 04/13/2009, -0/+2"Normal hard drives seem so archaic. "
In every way except one --- cost. - inactive, on 04/14/2009, -0/+2You know they keep telling us this. But where is my flying nuclear car. That's what I want to know damnit. It's always 10 years out, 10 years out. Well I'm starting to think it's not coming.
- Tenoq, on 04/14/2009, -0/+1Media Centre is the best way to kill any HDD - SSD or otherwise. The way it constantly writes to the disk when watching Live TV (even when not recording) cannot be a good thing!
- pak314, on 04/13/2009, -0/+1Briefly MLC is 4x slower write performance and 10x lower endurance than SLC but about half the cost.
- Hello1024, on 04/13/2009, -0/+1Yep, I agree some flash translation methods can't be represented by my very simple function prototype in the grandparent post, but that doesn't mean it wouldn't still be possible to implement them in a high level language as a prototype for such a "competition".
For example, the task could be defined as follows:
********************************
Your task is to produce a computer simulation with two interfaces as defined below. The simulation need not run in real-time, but will be scored according to "virtual time".
PC interface:
The system will be given asynchronous read or write commands consisting of a start address and length at particular times. The address shall be in a range which is 90% of the total number of bytes of flash memory.
The system must return data to a read command or a signal guaranteeing data is written to every write command. The time between the command being issued and the response is the latency. The average latency for the defined set of commands shall give the score - lower is better. This interface must have fewer than 1 in 10^15 bit errors.
Flash interface:
The system shall independantly control 16 flash interfaces. Each flash interface may read, write, or erase at any point in time. All operations take 100ns. reads and writes operate on 4kb blocks, erases work on 128kb blocks. Blocks may be written more than once, but each bit may only change state a maximum of once. Every bit has a 1 in 10^10 chance of failure after every write operation. A failed bit returns an undefined value. The device may operate flash when it has no pending PC interface jobs (eg. for garbage collection or consolidating data)
The system may store in volatile memory up to 128MB of data. The system must be able to store all written data that has been acknowledged on the PC interface within 100 µs in case of a power failure.
Total score shall be dependent on number of write operations before being unable to satisfy a write due to excess failed blocks, together with average latency for a selection of workloads. (by measuring latency this way we effectively also measure bulk throughput - consider a single read command for huge amounts of data)
********************************
With a relatively simple spec like this, most good algorithms can shine, and the spec allows many things such as compression, advanced error correction, automatic read and write profiling at runtime to optimize block placement, background "tidy up" tasks, write combining, out of order execution (eg. Native command queuing), read and write caches, a bad block list, incremental writes, a log filesystem, data striping or mirroring etc. Many things the students come up with might not be practical to implement in hardware, but as you suggest, many algorithms are implementable with an FPGA, and could be made in custom logic for mass production adding minimal cost. - Topher06, on 04/14/2009, -0/+1I will get SSD along with my $100 WiMax enabled laptop with 24" fold out OLED screen.
- xxmjumpman23xx, on 04/13/2009, -1/+2We need SSDs asap. Normal hard drives seem so archaic.
- Hello1024, on 04/13/2009, -0/+12. very few large companies would risk directly copying a competitors source code, and even without source code it's relatively easy to copy ideas by reverse engineering hardware or firmware, or even by simply analyzing the inputs and outputs of a piece of software. (eg. which bits of data get placed where on the flash) For those reasons I don't believe a "leak" of some or all of the algorithm would be a significant set-back for the company, particularly since any potential competitor would be 6 months later to the market with any "copied" ideas.
I believe the reason companies keep source code so private is mostly legal patent worries - a company might accidentally violate huge numbers of software patents, but if the software is closed source it becomes much harder for a patent holder to sue for use of some obscure algorithm, because they can't prove it's in use. - toastgodsupreme, on 04/13/2009, -0/+11. That makes too much sense. So obviously a company can't do it.
2. Having a competition of that sort leads to a very probable leak of the firmware code either intentionally by the winner or accidentally. Which then pretty much kills most of your advantage over competitors.
Honestly though, I'd love to see it happen. The competition idea with a very sweet prize for the most efficient algo is a great way to get some new ideas. And even if the most efficient algo isn't perfect, it'll still give the company's engineers things to work off of and maybe take them in a few directions they hadn't considered before. - morepowerr, on 04/13/2009, -2/+3ya ya just put it on the market all ready damn slow ass tech.
- WhaneTheWhip, on 04/13/2009, -4/+4All your rotary based HD are belong to SSD.
- benow, on 04/13/2009, -0/+0More than large enough for what an SSD is best for. Throw the OS and frequently used apps and dbs on it, and you'll have space to spare. Longer term archival should still be on spinning disk... there's no need for the low latency of ssd for stuff that's used very infrequently. Only apps I can think of offhand that require large SSD is multi-user multi-video unicast video streaming and HD video editing.
I do quite nicely with a 16G fast mtron ssd (7500 pro). The os (ubuntu linux) and dbs sit on the SSD and the rest is on std hd and symlinked into the filesystem. With lots of ram, it's even faster. The intel SSD X25-E is the nicest of the nice, currently, afaict. - benow, on 04/13/2009, -0/+0Well, VM's would be nicer on SSDs for sure.
- AlKo, on 04/13/2009, -1/+1It will be interesting to see what how this new firmware handles random writes... of the larger problems for NAND-based SSD. If it's something truly revolutionary, or something that's already been done by other companies.
- kr1srobn, on 04/13/2009, -0/+0Please read article before posting comments.
The point of the article is that SLC is faster and more durable, but also more expensive. With the new controller they are promoting, MLC could be nearly as fast, and just as durable at a lower cost. - benow, on 04/13/2009, -0/+0Probably because the flash addressing depends on the implementation. It's a good place to optimize, for sure. Violin SSD's use a binary subdivision algo to resolve bits across memory banks (and a FPGA for each memory back for addressing), which provides 10Gb/s sustained. It's a wee bit more expensive than even MLC ;)
http://www.violin-memory.com/Violin_1010 - Megadeth222, on 04/13/2009, -1/+1Yeah its cool and all, but it still cant run crysis........oh wait, *****....nvm
- aegis17, on 04/13/2009, -1/+0Anyone else notice the Steam icons on that computer's desktop?
Looks like someone's been doing a bit of "benchmarking" of their own... - AmericanGunner, on 04/13/2009, -12/+6New Solid Snake?
© Digg Inc. 2009
— Content created and posted by Digg users is