Solid State Drives - Fast, Rugged, and Expensive

Nick Breen writes "Are solid state drives becoming a reality? Loyd Case over at ExtremeTech has written an article concerning the current state of SSD with a comparison between a Samsung 64GB SATA and a Super Talent 32GB SATA. While they showed impressive speed rates when placed against a hard disk drive, the occasional sporadic statistic (and high cost) indicate they're not quite ready for the mainstream. Dell and Alienware have been shipping laptops with SSDs for months now, and Apple may be rolling out one of their own next year. Is the time of the solid-state drive almost at hand? Does anyone have any first-hand, practical experience with SSD?"

Huh?

[ScrewMaster (602015)]

What exactly is a "sporadic statistic"?

Re:Huh?

[Tackhead (54550)]

> What exactly is a "sporadic statistic"?

A statistic that is neither a lie nor a damn lie.

They appear very sporadically. (For values of "sporadically" approaching epsilon, at least 19 times out of 20)

Re:Huh?

[Aehgts (972561)]

A statistic that is neither a lie nor a damn lie.

Ah, so like a drop-bear [slashdot.org]. Scary, but thankfully nonexistant.

Re:Mods, Bad Mods, and Current Moderators

[Mountaineer1024 (1024367)]

Funny should be given a positive score, I personally consider a funny comment at least as valid in the conversation as an insightful or informative, and the current scoring system just leads to mislabeled posts.

the executor

[User 956 (568564)]

Does anyone have any first-hand, practical experience with SSD?

I know Darth Vader had his own SSD, but that's probably not what you're talking about.

Re:the executor

[Tavor (845700)]

Yeah, but the seek time on the Executor is horrible! Ever try to find one worker on the port side, when you are on the bridge? Not to mention the random A-Wing events causing the whole drive to crash!

Got one, love it

[by Anonymous Coward]

I've put one 32GB 1.8" IDE SuperTalent SSD in my Thinkpad X40, to replace that ever-failing 1.8" mechanical Hitachi crap, and formatted it with Reiser4 + cryptcompress. I LOVE IT. Fast, silent, more battery life, and, best of all, reliable. It was worth every buck.

Re:

[calebt3 (1098475)]

Since these drives do not have a head moving along a platter, what would be the most efficient partition format for them? It's not like the same rules apply.

Re:Got one, love it

[Abalamahalamatandra (639919)]

That doesn't sound right to me - I believe, unless I'm mistaken, that the controller on the drive levels writes across the entire drive, regardless of the partitioning scheme in place.

So even if your drive has, say, four partitions and one is written to a lot more than the others, that doesn't matter because the controller considers the entire flash space for write leveling.

Where is this applicable?

[RiotXIX (230569)]

I was going to buy a small one (15GB?) and put my linux partition on it (PC, so mobile benefits don't matter), but figured not too because of the fact that the number of times you can re-write is less. But according to "Because of these wear-leveling techniques, and the fact that a modern NAND device can sustain up to one million write cycles, the overall lifetime of an SSD can be decades. So losing capacity due to flash write cycles is probably not an issue", the option is now still back on.

But the re-write times are twice as slow! (ok I can live with that). But the read times are faster...as a home user, WHERE is this going to benefit me? Will I notice a diffence in 'vim file' or playing/streaming music?

I could maybe see if I were using a laptop, but I don't get how this would benefit me.

Thanks for taking the time to answer if anyone can persuade me different.
I might just get it for the cleanness of having a small segregated linux drive - really that's the best reason I can see.

Re:Where is this applicable?

[beavis88 (25983)]

For me personally, the biggest benefit would be silence. My hard drives are easily the loudest part of my machines.

Re:

[zippthorne (748122)]

Seriously? Not your power supply fan or heat sink fan?

Re:

[beavis88 (25983)]

Nah, not even close. All my fans are 120mm and run at 7V, making them pretty much inaudible. My power supply is a Seasonic S12 with an extremely quiet 120mm fan as well. It'll ramp up fan speed as temperature increases, but even with four disks and an 8800GT I've never been able to hear it above the disk whine.

Re:

[Propaganda13 (312548)]

Performance wise, once you switch to desktops, you are able to use performance drives like Western Digital Raptor WD1500.

http://www.storagereview.com/php/benchmark/bench_sort.php [storagereview.com]

Compare the Western Digital Raptor WD1500 No NCQ to the Western Digital Scorpio WD2500BEVS with NCQ (250 GB SATA). The Scorpio consumes a lot less power, but isn't that much quieter. The Raptor has about 2.5x the performance.

SSD wins on noise and power, and the Raptor wins on price. Depending on the application, either could win in

Re:Where is this applicable?

[tlhIngan (30335)]

But the re-write times are twice as slow! (ok I can live with that). But the read times are faster...as a home user, WHERE is this going to benefit me? Will I notice a diffence in 'vim file' or playing/streaming music?

Actually, if you do any sort of multitasking, you'll probably notice it's a lot "snappier" (apps load faster, switching apps doesn't seem to take so long, etc). Or if you're a typical home user with decent RAM but still have all the usual crapware loaded, WIndows won't feel so slow. Or you don't defragment your disks and let your disk get horribly fragmented...

The deal with SSDs is that they can manage their peak datarates all the time. With disks, the smaller the I/O transfer, the slower the disk becomes. If you have a disk with a 5ms seek time, you're limited to 20 I/Os per second. If you read maybe 16 sectors each (8kiB), it means your disk throughput is on the order of... 160kiB/sec. Seeks are taking a lot of time compared to the actual time it takes to read the disk.

An SSD has negligible seek time, so reading those 160kiB off an SSD won't take noticably longer than reading 160kiB in one read (the overhead of doing the transaction over the ATA bus is the biggest overhead).

You won't use an SSD if you need high throughput, where you're basically doing huge writes or huge reads (i.e., media center media disks, video capture/production, etc). But a home user that's doing a lot of little random I/O will notice that the entire system feels "snappier" as the I/O is mostly seek-bound, not throughput-bound (small I/O). This applies as time goes on as most people don't defragment their disks (you don't have to, or should, with an SSD, since wear-levelling may still not put it contiguously on the flash media), so even a heavily fragmented disk will still feel fast with an SSD.

Re:Where is this applicable?

[afidel (530433)]

You're off by a factor of 10 there, 1sec/5ms=200 I/O's per second which still gives only 1.6MB/s for totally random reads for 8KB blocks.

Re:Where is this applicable?

[timeOday (582209)]

WHERE is this going to benefit me?

Did you look at the "real world" benchmark results [extremetech.com]? The Samsung SSD drive destroyed the traditional drive by 400%-500% in 6 tests (including OS startup, app loading, gaming) and was about equal in the other two (media center and video editing).

Unless you know of some special reason why sustained write speed is critical, you should probably be looking more closely at access time, where SSD blows mechanical drives out of the water.

No doubt, mechanical drives still rule capacity/price, but with the growth rates of the two technologies over the past several years, SSD could take over soon.

Re:

[timeOday (582209)]

Once again: "seek time."

Re:

[GoofyBoy (44399)]

Database Servers.

I use them

[Lord Ender (156273)]

Both my home server and several systems in use at work boot from compactflash drives. Our production servers run Ubuntu LTS, and are basically VMware Server boxes--the actual apps run off of guest OSs that live on the 6TB RAID-6s on each server.

All in all, I've had seven servers running off of SSDs for about eight months, and they have worked like a charm. I never have to worry about getting paged due to the inevitable mechanical failure of magnetic drives.

Also, SSDs are NOT expensive! A CF-to-IDE adapter costs $15, and a 2GB CF card costs about $30. Two gigabytes is more than enough to boot an OS and start a RAID. Don't waste your money on a 64GB CF card. The CF+RAID hybrid approach is the way to go.

Re:

[jon287 (977520)]

I do the same trick. I've had more than 300 systems running for around 3.5 years, both read only and read write with plain old ext3. Not a single cf card has failed. Nearly 1/5 of all the rotating disks we've delployed in the same time frame have either failed completely of shown some sort of strange behavior or smartd error. Booting off the cf card leaves us with enough system after the mechanical disk fails to tell us what has gone wrong without an expensive truck roll.

It is true that eventually cf cards

laptops, dummy

[DreadSpoon (653424)]

There's a reason that these things are commercialls available only in laptops right now. In a laptop, you boot up a lot (or resume from hibernation a lot, which is equally disk-intensive), so disk seek and read times are incredibly important. Plus, power savings are a huge benefit when you're running a system that has a limited power source. The SSDs generate less heat, which is also hugely important when all your circuitry is compacted into the smallest amount of space possibility, and it allows either for the system to be cooler (hot laptops suck, even typing on them can be uncomfortable) or allow for other components like the CPU and RAM to be sped up since they get a greater share of the system's safe heat generation capacity. The reduced noise is great - try being in a meeting with 20 laptops all with fans whirring away. Finally, the greater lifetime of an SSD (modern hard disks fails way sooner than a modern SSD will, in general) means that the machine doesn't need a new disk with a new OS install and possibly a bunch of lost data on anywhere near as frequent a basis.

Less power and less noise are good for servers and desktops, and the faster seek times can really make a different in performance for many common workloads, but the biggest benefit of SSD is that they make laptops suck way less.

Re:

[Lord Ender (156273)]

Well, the CF fits directly into the motherboard, freeing up two more slots in the case for the RAID. Also, we spec'd the servers and ordered them from a local business, which built them for us. Buying through our "official" lines would have cost quadruple the price. It cost us no extra time because nobody would have sold us a preconfigured Ubuntu LTS server anyway... at least not with the kind of hardware we required.

There's also the inherent awesomeness of booting from flash.

First-hand, practical expericence...

[What the Frag (951841)]

> Does anyone have any first-hand, practical experience with SSD?
Yes. Transcendent 4GB 266x Compact Flash card, fast, silent, installed Ubuntu 7.04, currently 1.4 GB free.
Price for the card + card to ide bridge was about two 80GB HDD drives.

Only problem was that I had to make my own drive mount first, because all I got was a board with a Compact Flash slot and a IDE connector.

If you are happy with a few GB of disk space, go for it. If you want to store big amount of data, wait. The price will fall.

similar storage, different form factor

[Corf (145778)]

I am currently typing this on one of them newfangled Asus Eee PCs. 4gb worth of Hynix HY27UG088G5M chips through a Silicon Motion SM223 controller. The only moving parts on this thing are the keys and this near-worthless little sideways-blowing fan. It's fast, reliable, shock-resistant, and pretty durn cheap.

Specs [eeeuser.com].

Like Digital Cameras

[JonathanR (852748)]

I think solid state drives will be like digital cameras. The price and usability (read size) will appear not to be mainstream enough, that is, until you've just made that "big" investment in the latest incarnation of the superseded technology.

It happened to me. I bought a new (not that expensive) film SLR about 18 months prior to digital cameras having sufficient resolution/cost ratio to supersede film for everyday use. Coming from a generation where cameras tend to last almost a lifetime (having been used to my father's Minolta SR-T 101, purchased about the time I was born). The concept of a camera becoming almost obsolete in that short timeframe was a bit annoying, at the time.

First hand

[by Anonymous Coward]

I have first hand experience with SSDs as I have bought one of the Samsung 64 GB SATA SSDs. In terms of writing performance, they're approximately on par with regular hard disks, as far as I can tell. Disk reads, however, are very good. To give you a vague idea of the read speed, Windows XP on this drive boots to login screen without the black logo screen appearing at all. Additionally, for those who are interested, here's what Linux's hdparm has to say about it:

# hdparm -tT /dev/sda1 /dev/sda1:
  Timing cached reads: 7352 MB in 2.00 seconds = 3679.72 MB/sec
  Timing buffered disk reads: 168 MB in 3.01 seconds = 55.86 MB/sec

Re:First hand

[mandolin (7248)]

I'm going to say that your first benchmark, at least, is completely fucked up.

No, those are cached reads, not hitting the drive at all. The man page for 'hdparm' says -T "is essentially an indication of the throughput of the processor, cache, and memory of the system under test".

It just boggles my mind...

[JustNiz (692889)]

It just boggles my mind how modern solid-state electronics organized for parallel I/O can be less than a factor of 10 times faster than an inherently serial and decidedly ancient-sounding "mechanically moved heads over a magnetized spinning disk" approach.
What the heck is going on here?

Re:

[Firehed (942385)]

That fails to explain how a drive made of, in effect, nothing but cache is about an order of magnitude slower. Whether you can process the data quickly enough is irrelavent when you're dealing with a medium theoretically limited by nothing but c yet performs worse than a device spinning at 4200RPM.

We're not considering the full system performance here. We're trying to figure out why something that has a seek time that's effectively zero isn't even maxing out the interface. A RAMDisk (those funny boards G

So far it's a mixed bag...

[BUL2294 (1081735)]

Basically the reviews on Anandtech [anandtech.com] & Tom's Hardware [tomshardware.com] have drawn some interesting conclusions... In terms of write performance, some are significantly worse than most notebook HDs, but all are better in terms of read performance. The idle of SATA SSD drives are significantly worse than UDMA ones (0.5w vs. 0.05w).

Basically, do your research... How much speed you'll get depends on how they bank the flash chips. More banks of lower density chips will yield a higher transfer rate--but uses more power. (Good luck finding how any one brand of SSD drive is banked...) Tom's Hardware found that the Samsung 64GB SSD offered double the transfer rate than their 32GB SSD. Anandtech found the Transcend & Super Talent SSD's to be extremely weak offerings. But then again Anandtech found the MTRON 32GB SSD far superior to most other drives they tested.

Basically SSD drives help with bootup times but in mixed tests, only the MTRON SSD drives are near Raptor speed, but I found only one retailer that even sells them--and a 32GB one for $2336.95 [google.com]!!!

SSDs

[phoophy (1189235)]

Been using arrays of 4 and 8 32GB SSDs as both RAID0 and RAID5, off hardware RAID controllers and as Linux softraid, to push seek time to near 0 and throughput as far as possible. Bottom line is, they're significantly faster than "real" disks. We've found MTrons to be faster than Samsungs, generally 20 to 40%, and the MTron seek times are significantly better (they probably don't write-balance check as often under heavy usage). Only reliability problems I had were with another brand (neither Samsung nor MTron).

Give me RAM

[evilviper (135110)]

All I have to say is screw NAND. Give me some DDR RAM-based hard drive... Ridiculously fast, very low power, no possible questions about lifetime. Perhaps even the possibility of just swapping out one failed SODIMM instead of scrapping the whole drive, is quite enticing.

I've been using Flash longer than most... From wiring minuscule capacity EEPROMs into embedded circuits, to squeezing OSes down to 8MBs for firewalls. Floppies are a no-go for important systems.

They're low power, quiet, and have high speed seeking, but I don't really care. What I want most in a drive is seriously high throughput... That probably means RAM, with a battery back-up. In the mean time, HDDs keep getting faster and quieter.

Data recovery from SSDs?

[rHBa (976986)]

I was wondering if anyone can answer a simple(?) question: Can data be recovered from an SSD after it has been overwritten once? i.e If I'm disposing of an SSD with sensitive data on it do I have to run secure erasing software to make multiple/random writes to every sector?

I have an EEE with a solid state disk drive

[Tracy Reed (3563)]

And I really like it. This laptop is great. I have a desktop with dual 24" displays for doing work so I don't need a laptop for that. What I do need is something ultra-portable to do email, read slashdot, occasional ssh into a remote machine while on the road, terminal into a box while at the datacenter, etc. And this thing fits the bill. The solid state disk has caused no problems so far but allows things like 10 second boot times and no noise and little heat. The prices of SSD will come down, the densities will go up, and SSD drives will proliferate.

3 months real-word experience with SSD

[barre (48638)]

I've been using a Sandisk 32 GB SSD on a Dell Latitude D630 running Vista for about 3 months now. This wasn't cheap, and even with an early adopter mindset, this is a big disappointment; it does indeed reads much faster (about 30 times), but writes at least 3 times slower than the same D630 running a SATA. My typical usage involved web/email, Microsoft Office, photography/photoshop, compiling large projects, etc.

Quiet is great, more battery is fine, and I hardly ever reboot using Vista almost instant-sleep feature, but installing software or writing large files is *painful*. Moreover, you should plan for a lot of physical memory: you do *not* want to see your system paging for virtual memory.

Now maybe Vista is to blame, but the whole system will hang now and then for 10 secs or more. Is it indexing something, writing whatever system logs to disk, who knows, but a a few other users have reported the same issue with this SSD on Dell forums. No driver update has been released either since the SSD option was out. This is also probably not coincidental that SSD vendors emphasize read speed but remain somehow quiet about the write speed (or lack thereof).

I, for one, am switching back to a 7200 RPM SATA. This is *not* ready for prime time, even if Samsung claims slightly better write speed on its 64 GB; *do* check the user forums (say, Dell), and you will find a lot of frustrated users. This was worth a shot, and I'll eventually consider that technology again in 10 months.

Hope this helps

Re:

[gillbates (106458)]

Now maybe Vista is to blame, but the whole system will hang now and then for 10 secs or more.

No, that's the Windows Vista Minus Pack - first introduced on Windows 98, the Minus Pack includes assorted features such as:

1. Hang Computer for 10 seconds for no reason. (Much improved since Windows 98, where it was only 2 seconds - a fivefold improvement!)
2. UAAM - Use All Available Memory. Get what you paid for! Use all of the memory you bought - Windows Vista will ensure that it uses all of your memory.
3. Manda

Puppy Linux

[flyingfsck (986395)]

Puppy Linux runs nicely on small USB memory sticks of 128MB and up. A 1GB memory stick make a beautiful system. You really don't need umpteen gazillion gigabytes of storage space for a PC.

I love my SSD!

[lumbricus (936846)]

I agree with TFA that SSD is most useful if you need the ruggedness and the read speed. I have a 16GB Samsung 1.8" SSD in my fujitsu P1510D. It's a marriage made in heaven! I am a Biologist, and use my P1510 in the field. The SSD gives piece of mind, one less thing to go wrong. In fact, almost right after I swapped in my SSD, I (yes accidentally) dropped my computer about three feet to the floor. After checking to see that screen and case wasn't cracked, I just knew it was fine, and of course it booted right up. I also work in some high altitude locations, and I find the the machine boots at higher altitudes now. (perhaps hard drives cut out at high altitude because there's not enough air to keep the head off the platter?) Finally, the P1510 uses hard to find and extremely expensive micro-DIMMs, so upgrading the memory is prohibitively expensive. That was my biggest gripe with the little machine, it was slow because I couldn't get the 1GB it really needs. This, coupled with the incredibly slow 1.8" hard drive made it kind of annoying. I still can't do much about upgrading RAM, but the read speed of the SSD allows me to just close applications, and re-open them when I need them (nearly instantaneously), so I never have more than two applications open at a time. The most telling test I've done is with Allway Sync, which I use to synchronize the files on my little laptop with my desktop. Running "Analyze" (version checking files) on my home folder used to take about about a minute, now with the SSD it's somewhere between 10-15 seconds. Sure, I wouldn't put it on a MySQL server or the like, but for my laptop, the whole experience is just so much better. I would recommend one to anyone who can use the ruggedness and read speed.

Is it?

[Frozen Void (831218)]

A year of Solid State Disk on Desktop?

Re:you left impractical off the list

[PoliTech (998983)]

"they seriously are not that much tougher then a laptop hd."

I would like to see a citation for that claim. From my team's research, SSDs are much much tougher than any spindle HD. But toughness may not be a factor for you when evaluating SSDs, (it wasn't for us).

Our test SSD laptops have also demonstrated much improved battery life. On a D630 we are seeing four and a half hour battery life with standard stock batteries. That's a two hour increase. Use larger cell count batteries and battery life will just get better. A laptop equiped with an eight cell battery and a secondary battery licated in the Optical drive bay, we have experienced eight hour-plus battery life.

Our boot times are also improved with SSD. Since we also encrypt, (and if anyone has used encryption on a Windows domain then they have likely experienced a hit with login times) we were most impressed with the performance improvement of encrypted SSD, when compared to a traditional HD on the same equipment. Write times are not as much improved, but there is no negative impact either.

Our experiences have been good enough that we are planning to order SSD on all new laptops for next year. The improvement in Battery life alone is worth the price of admission. Toughness, and increased write speed are icing on the cake.

Re:

[magarity (164372)]

On a D630 we are seeing four and a half hour battery life with standard stock batteries
 
I also have a D630 with the stock battery and I get just over 4 hours with a normal hard drive thanks to using RightMark's CPU utility to lower the voltage by 30% from the factory setting. You should try it in combination with the SSD and the thing will run for days on an extended battery. (your CPU might not be stable at 30% lower like mine but you'll be able to lower it some)

Re:

[timmarhy (659436)]

A little of both.

SSD's have a short life span due to cell memory, and they aren't immune to shocks damaging them. laptop hd's will take all kinds of poundings, only a direct solid hit during a r/w would possibly damage them

Re:

[HTH NE1 (675604)]

SSD's have a short life span due to cell memory

The larger the drive, the more spread out the wear, the longer it will ast. By my calculations [slashdot.org], a 1 GiB NAND Flash as a TiVo's video drive rerecording the same data every 30 minutes would last 570 years.

If I've made a mistake in those calculations, I'd appreciate a correction before I feel compelled to cite them again.

Re:

[timmarhy (659436)]

your assumptions are all flawed because you assume the best case scenario for everything.

I on the other hand am basing my assumptions on real world experiences working with industrial equipment that uses CF cards for hard drives in mobile fleet equipment.

in the real world i've seen a 10% failure rate on CF cards (which are tougher then SSD's i might add) over 12 months WITHOUT any write action at all.

Re:you left impractical off the list

[Sen.NullProcPntr (855073)]

If I've made a mistake in those calculations, I'd appreciate a correction before I feel compelled to cite them again.

Assuming NAND figures of 1,000,000 rewrites before a failure:

1,000,000 writes/bitfailure / 139.8 writes/year = 7153 years/bitfailure

I haven't confirmed the rest of your math but you appear to be off by an order of magnitude for the number of erase/write cycles without an error.

This quote is from a recent Intel 2Gb NAND chip; [intel.com]

First block (block address 00h): -- Guaranteed to be valid up to 1,000 PROGRAM/ERASE cycles

(you can view the first block as the boot block, that is - very important) And;

On-chip control logic automates PROGRAM and ERASE operations to maximize cycle endurance. ERASE/ PROGRAM endurance is specified at 100,000 cycles when using appropriate error correcting code (ECC) and error management.

I interpret this to mean 100,000 cycles without an uncorrectable error but you can expect to see random bit errors after only 1,000 cycles. You will need the overhead of error detection and correction as well as mapping the bad section of memory to another area (your read times will be slower than the theoretical max). You will find the 1,000/100,000 numbers pretty much standard among NAND manufactures.

That said, I agree that NAND is reliable and is most certainly _the_ replacement for mechanical hard drives.

Re:

[Sen.NullProcPntr (855073)]

My source for 1,000,000 writes before failure was Wikipedia, contemporaneous with the posting.

That was your first mistake;-)
Not sure where the Wiki is getting its numbers from maybe reference [5]? [dataio.com] an old (2003) Toshiba marketing pamphlet (for some reason hosted by a chip programmer company [Data-io]).
I would like to see a real datasheet claiming 1,000,000 writes.
Even Mtron is only claiming 140 years [mtron.net] for their SSD with its "advanced wear-leveling technology" (they reiterate 100,000 cycles for an individual chip).

Re:

[afidel (530433)]

and they aren't immune to shocks damaging them.

Yes they are, for all intents and purposes. If you don't believe me see this story [digitaljournalist.org] about a CF card that survived the collapse of the WTC.

Re:

[afidel (530433)]

Samsung's datasheet says their drive is rated to 1,000g, that's 10x better than even the best shock isolated laptop drives with physical spindles and enough shock that you'd probably break the motherboard, lcd, etc long before you damaged the drive.

Re:And, the MTBF is..

[BlueParrot (965239)]

While storing it as an ultra-small magnetic dipole moment in a piece of rust on a rapidly spinning platter which will be irreversibly damaged from just a speck of dust sounds like a sane idea ?