2008, The Year of Solid State Storage

An anonymous reader writes "At CES, SSD drives were a plenty on the show floor. "Some companies said we could see 250GB SSD units by the end of this year, while others predicted it could take up to a couple of years for them to become mainstream. None of the companies promised mainstream adoption, but they promised a bright future and we are inclined to believe them. High capacity drives are going to be expensive due to their very nature of early technology and gradual adoption rate."

Sequential reading?

[ookabooka]

I was talking to a gentleman from (big name hard drive company) about their plans for hybrid and/or solid state drives. Essentially he told me that solid state was still limited by price and sequential reading. So it may be advantageous to put some things on flash like OS files that require a lot of random seeks, but for sequential reading of things like media files, traditional hard drive tech won't die just yet . . .I apologize for being too lazy to back this stuff up with numbers, what can I say, I'm a true slahsdotter.

Re:within 5 years, tape manufactuers will have tro

[Anonymous Coward]

That's hardly the situation at all. For massive magnetic data storage, tape is still very valid. You're just not going to find 500GB HDD's with such low failure rates in 10-packs for $1000 like you can get tapes at today. And tape can drop in price much more easily than HDD's will.

I'd give it a good 10-15 years before our massive tape storage units disappear from the datacenters.

Re:Sequential reading?

[jdunn14]

I've never seen the performance numbers for sequential vs random read on flash drives, but you have to do pretty damn bad to get beat by random access on a standard hard drive. If you look at the the units used you'll get the idea. Your average random access on a standard drive is based on the average seek time which is measured in small milliseconds (4 ms, 8ms). Access time for flash drives is measured in double-digit nanoseconds (e.g. 60ns). That's 5 orders of magnitude difference. Even if the access time for random reads on flash was 100 times worse than it's average access time those reads would STILL be 1000 times faster than from a hard drive.

I don't think people realize just HOW slow drives are compared to the rest of the machine. Sure we programmers know the disk is "slow" but it really puts it in perspective to know it's a 100000 times slower than an alternative tech.

Reports I Continue to Hear

[Nom du Keyboard]

Reports I continue to hear is of blocks going bad (meaning that overall storage is reduced by measurable chunks, rather than failing all at once the way a head-crash on rotating media can happen) in as short as weeks of use. Especially when the drive is rather full to start with, since wear leveling doesn't tend to move stored data to empty slots.

Until that time is years, instead of weeks, I don't see myself preferring more expensive, or even equal cost SSD, over rotating media drives.

Re:Lets try the other way around, eh

[Geoffrey.landis]

Price isn't the only factor here. Has anyone seen any real reliability or Environmental numbers on any of these drives yet? I know many government/military programs who would be glad to pay for it, if it could prove to increase availability in certain environments.

Well, flash storage certainly is better in the space environment. Conventional hard-disk technology requires a pressurized compartment (the heads stay separted from the disks with a thin film of air). And, of course, any technology with no moving parts is preferable-- mechanical parts have an annoying tendency to freeze up with vacuum thermal cycling.

Spirit and Opportunity are now four years into their 90-day mission on Mars, running on flash storage....

It's not just for laptops...

[johnmcd]

Looks like the big boys are getting into the game also: EMC in Major Storage Performance Breakthrough; First with Enterprise-Ready Solid State Flash Drive Technology Market-leading Symmetrix DMX Systems to Feature Newest Flash-based Technology for Unprecedented Performance and Energy Efficiency http://www.emc.com/about/news/press/us/2008/011408-1.htm [emc.com] They're claiming a 10X performance improvement, but at 30X the cost/MB. Given that a high-end DMX holds around 3000 drives, that a lot of flash memory! John

Re:Lets try the other way around, eh

[Lars Clausen]

Just looking at newegg.com, I find the current sweet spot of SSD to be 32GB, at $250. $7/GB, half down from what Wikipedia mentions for "late 2007". The price is not just falling, it's plummeting like a jumbojet with both wings shot off. I love it. Can't wait to get the last mechanical pieces out of my computers.

-Lars

Re:So we are back to RAM drives!

[Sen.NullProcPntr]

I wonder when we might see a hybrid flash-ram drive? A big bunch of ram for high speed and flash for permanent storage.

Normal magnetic hard drives already do this [wikipedia.org] to speed up sequential access (read ahead) among other reasons. No reason to believe this feature won't be transfered to SSD media. Although flash is much faster than magnetic media already.

EMC Solid Storage Array just anounced.

[BrianHursey]

EMC just announced this this morning. They are going to start haveing soldi state drives in there DMX-4 storage arrays. "EMC plans to offer flash drives in 73 GB and 146 GB capacities for the Symmetrix DMX-4 platform beginning later in Q1 2008". http://www.pcworld.com/businesscenter/article/141323/emc_readies_solidstate_drives_to_replace_disk_storage.html [pcworld.com]

Re:Lets try the other way around, eh

[mlush]

No, no.
Hard drives are actually vented. There's no pressurized compartment. They run at the same atmosphere as the rest of the machine. The lift of the hard drive heads is the "Bernoulli effect" (http://en.wikipedia.org/wiki/Bernoulli's_equation) see also (http://news.bbc.co.uk/1/hi/programmes/click_online/5413198.stm).

So when these drives are exposed to hard vacuum (as suggested by the OP) the Bernoulli effect fails and the heads start gouging into the platters.

Flat panel/CRTs all over again

[JerryQ]

In (approx) 1992 I went to SID (society for information display) in Florida, and a keynote speaker said, roughly: "I have been coming here for 30 years, and I expect to hear, just like I heard 30 years ago and most years since, that within 10 years flat panels will overtake CRTs and make them redundant. Why has this not happened? Because CRT has continued to get cheaper and better quality, thus removing the opportunity for flat panel, because the goals keep moving" He also pointed out that we would get there (and we have) but that we should never underestimate where old technologies can go. In 1983 I put together a business plan for an outsourced proposal I was working on, and we put in £17K (thats $28k) to cover a 70 megabyte hard drive. Now I see one inch drives in iPods carrying multi gigs. I believe that we will see phased take up, ie where it is needed most (e.g. like the way airlines put in flat panels instead of CRTs to reduce weight), before the HDD manufacturers will curl up and leave the scene. Jerry

Re:Sequential reading?

[kestasjk]

Well those are seek times, as you said. Reading/writing continuous data is very fast, and the OS (and some HDDs) will use memory caches so that data access will be continuous as possible. The problem of hard disk seek times has become less and less of a problem as memory has increased.

Re:I dont see it

[initdeep]

I have a 64GB SSD Drive in my Dell M1330 Laptop. I too use an external as storage for files.
The differences, side-by-side, to one without it, simply for OS Startup, are easily 3 to 1 in speed.

I was lucky enough to basically get the drive for free due to the EPP program coupons and discounts and other discounts..
otherwise i would never have gotten it.

But I'm sure glad i did.

I've also noticed a slight increase in battery life, although this could be simply a small difference in batteries themselves.

Re:SSD as a boot drive

[Heddahenrik]

One thing that has been tested successfully in various places is to RAID (RAID0) the SSD-disks. It makes them about as twice as fast, and it should be possible to RAID them in bigger arrays too. As there is so little risk of one disk breaking down, there is no excuse to not RAID-0 them.

These disks still have a problem with speed on random write though. It's nothing for read-write databases where NCQ (SATA2) disks are faster.

Re:What about limited write cycles?

[Wolvie MkM]

Here>/a>, there's a mention about file systems that will accomadate the problem that SSD drives have. Basically spreading out the bits all over the drive if my interpretation is correct. [wikipedia.org]

As for XP, you don't "need" a page file if you've got over a Gig and don't do anything that will come close to the limit. I've removed the page file on all of my systems and been fine (3 Gig Desktop, 2Gig Desktop, & 1Gig notebook). It's an easy way to save some space if you know your limits!

Cheers hope that helped

Re:Lets try the other way around, eh

[amorsen]

Bulk trasfer speed does not have to be slow. Its not like your hard drive in your PC now reads from one head at a time.

Err yes, in fact, that's exactly what it does.

It uses multiple heads to achieve higher rates by reading/writing all the platters at one time.

No. It was attempted, I believe by Seagate in the first Barracuda drives, but it was quickly abandoned. The only way it can work at modern capacities is if you added a drive motor and independent electronics per head. Doable, but it's cheaper to just buy two drives and do RAID-1.

Your points about flash are correct though.

Re:CDMA works for hard drives too!

[BillBrasky]

Hard disk drives defintely do NOT write simple binary waveforms to the disk. They use encodings specialized for magnetic media, such as Extended PRML [wikipedia.org]. This is coupled with error correcting algorithms like Viterbi (same as CDMA).

Re:Lets try the other way around, eh

[dfn_deux]

The 32GB SSD you can buy for 250USD will be unlikely to provide any performance or reliability increase over what that same 250gb would buy you in traditional magnetic media. I've done perf/reliability testing on all the current generation SSDs from mtron, stec, memoright, and ritek. And even the 800-1000USD SSD drives fall way short of their predicted write lifetime when put into any environment where i/o is primarily small random r/w operations vs large sequential stream r/w operations. Write leveling is generally limited to 512b or larger blocks which makes it ineffective for these types of loads, furthermore the cheaper devices tend to use MLC flash nand vs. the less dense SLC.