Slashdot Mirror

you would think the amount of cord-cutting would get more multicast video streaming on the net....

Long haul bandwidth is expensive but caching is cheap and the last mile cost mostly installation and maintenance, so there's not much savings from multicast over unicast from a CDN. If it wasn't for copyright you could set up a super effective simple, open LRU cache for torrents, every time you download externally the ISP caches the chunks. If one chunk = 4MB then a 4TB HDD = a million chunks and you don't need any redundancy. Next person who downloads only sends the hash and gets it right from the ISP's network at max speed. You could even do multi-layer and have a half-PB storage pod before you use any expensive links like undersea cables. For uhh... Linux distributions, sure.

An off-the-shelf SATA 840 EVO SDD hits 98,000 read IOPS

Well, at first, anyway...

On a PC environment when you've got multiple browser windows open, IRC, email client, etc. getting constrained for IOPS is easier than expected.

An off-the-shelf SATA 840 EVO SDD hits 98,000 read IOPS, and all those tasks you mention added together wouldn't hit more than 1% of that. They're the very definition of network bound operations. The average email in my IMAP spool right now is 43KB and would take 11 4KB operations to completely read from or write to storage. Browsers site there idle 99.9% of the time. IRC? Not that I've ever seen.

Do it in a real world environment, and I'm willing to bet PCIe will show it's worth. I don't think that games will run any faster than the baseline results of no load, but I'm willing to guess it'll do better than the SATA equivalents.

I haven't bothered to look at their methodology but I tentatively agree with their conclusion: almost no desktop users would be able to tell the difference. I mean, even a HDD benching at 103 read IOPS seems spritely for most use cases. A SATA SSD working 950 times faster is as close to instantaneous as most desktop uses could ever hope for.

On a PC environment when you've got multiple browser windows open, IRC, email client, etc. getting constrained for IOPS is easier than expected.

An off-the-shelf SATA 840 EVO SDD hits 98,000 read IOPS, and all those tasks you mention added together wouldn't hit more than 1% of that. They're the very definition of network bound operations. The average email in my IMAP spool right now is 43KB and would take 11 4KB operations to completely read from or write to storage. Browsers site there idle 99.9% of the time. IRC? Not that I've ever seen.

Do it in a real world environment, and I'm willing to bet PCIe will show it's worth. I don't think that games will run any faster than the baseline results of no load, but I'm willing to guess it'll do better than the SATA equivalents.

I haven't bothered to look at their methodology but I tentatively agree with their conclusion: almost no desktop users would be able to tell the difference. I mean, even a HDD benching at 103 read IOPS seems spritely for most use cases. A SATA SSD working 950 times faster is as close to instantaneous as most desktop uses could ever hope for.

You're assuming one slot, but they actually sell multiple slot monstrosities that are aggregated together as a single drive. Here's a similar one to what I tested benchmarked at 5.8GB/s on reads.

Well, so do HGST and Toshiba desktop drives, so the distinction is not quite as clear cut...

Do you have a reference for this? This information doesn't seem trivial to find, the closest I can find is "The Deskstar NAS also offers configurable advanced error recovery control to fine-tune RAID performance." in a review of the HGST 4TB Deskstar NAS HDD and no such claim in the review of their non-NAS drives. Do you have a reference showing that HGST supports TLER/ERC/CCTL across their desktop (non-NAS) drive line? I don't think Toshiba has a NAS drive though I'm not very familiar with their product line.

Let's do some math here, shall we? At 200 MB/s, you can overwrite a 1 TB drive in an hour. 1 PB you can reach in a month. The hard drives are a few times larger than the SSDs, so you'd need ~ 10 TB instead of 2, which means 10 months.

Include all the actual variables, and you might get a usable answer. Just blowing data on the disk isn't the only thing this is doing (AFAIK). You've gotta detect errors, so you've gotta read back the data and validate it. This page goes through their full testing methodology (hint: they're using Anvil, a static file collection that includes a copy of a windows install, some applications, some movies, and some incompressible data, among other things, and every file has its md5sum checked after writing): http://techreport.com/review/2...

An easier calculation would be to scale their timelines to the HDD stats. For example:
Samsung 840 Pro sequential read/write: 540MB/s / 520MB/s (390MB/s for 128GB)
WD Caviar Black: about 180MB/s read/write (ex. http://www.storagereview.com/w...)
Rough math: 520 / 180 = 2.89 = it'll take 2.89 times as long to do the test on the same size drive.

Samsung 840 Pro size in the article: 256GB
Assuming you use WD Caviar Black 1TB = 4x's the size.
2.89 * 4 = 11.55 = that many times as long to do the same operations they've done thus far.

Their test has been running for over a year. So it'd take (roughly) over 11.5 years to do the same on the WD Caviar Black. I understand that's a very very rough estimate, but I think it's MUCH closer to the ballpark than 10 months!

My bet: the WD will be dead long before that time. I've had drives last longer than that, but they got VERY VERY little use and were simply powered on all the time. I've had some that lasted longer than that and got a fair bit of use (ex. db servers), but they were never filled to capacity, they were enterprise drives, and some of their neighbors did die (RAID).

Recording TV is not a typical scenario. Besides, at around 8GB/hour (HD), that's around 2000 hours a year, which is little more than what my BeyondTV machine does, and its 3TB WD green is still alive and kicking. You just have to disable the insanely aggressive head parking on those drives otherwise they might die...

http://www.storagereview.com/h...

Not that much higher for streaming reads and writes, the new Seagate 6TB can do 220MB/s @128KB streaming reads or writes. That works out to ~19TB/day so it would only take around 2 months to hit 1PB.

From a review of the Samsung 840 EVO 1TB SSD I just stuck in my MacBook Pro:

• Sequential READ: up to 540 MB/s
• Sequential WRITE: up to 520 MB/s
• Random READ: up to 98,000 IOPS
• Random WRITE: up to 90,000 IOPS

From the same site reviewing a WD Black 4TB HDD:

Performance from the WD Black scaled from 66 IOPS at 2T/2Q to 86 IOPS at 16T/16Q, versus the 7K4000 which scaled from 82 IOPS to 102 IOPS.

So assuming IOPS scales linearly with heads (they don't), you'd need about 1,000 heads to get similar random access performance out of HDDs as one SSD.

There's a reason everyone's migrating to SSDs for anything remotely IO related.

From a review of the Samsung 840 EVO 1TB SSD I just stuck in my MacBook Pro:

• Sequential READ: up to 540 MB/s
• Sequential WRITE: up to 520 MB/s
• Random READ: up to 98,000 IOPS
• Random WRITE: up to 90,000 IOPS

From the same site reviewing a WD Black 4TB HDD:

Performance from the WD Black scaled from 66 IOPS at 2T/2Q to 86 IOPS at 16T/16Q, versus the 7K4000 which scaled from 82 IOPS to 102 IOPS.

So assuming IOPS scales linearly with heads (they don't), you'd need about 1,000 heads to get similar random access performance out of HDDs as one SSD.

There's a reason everyone's migrating to SSDs for anything remotely IO related.

If you RTFA, they break down the failure rates by model (no pun intended). There's a pretty huge variation between models (or at least the Seagate models). That's also what I saw in the StorageReview reliability database back when people were actively updating it (unfortunately you have to add a drive to the database to get access to it, so it was never very popular). The same manufacturer can make a gem and a stinker of a model. e.g. the IBM 75GXP (aka Deathstar) drives had one of the highest failure rates in the database. The drive which replaced it (60GXP I think) had one of the lowest failure rates in the database.

So it's more nuanced than "Seagate stinks, Hitachi rules." (Hitachi is a subsidiary of WD now, operating separately only because that was a condition China placed on them before they'd OK the merger.)

The write cache on the Crucial M500 survived my power plug testing with similar properties to this article, but it doesn't have SMART data for drive wear. Can't take that one seriously for business use. I've been very happy with the figures I've seen from Intel drives on their internal lifespan tracking; see my Intel SSD lifespan for example. I'm not the only one who noticed this flaw in the M500, the Tech Report review has another complaint.

None of the cheap and easy to buy Samsung drives (840 and 840 Pro) claim any power loss protection, and they all fail this sort of test. They have enterprise models that might work, but those wouldn't fit within the budget parameters here. I find it hard to take those seriously when the 840 drive they share design features with are so terrible handling lifespan failures.

I'm not aware of any Sandisk models with that feature, but I haven't gone looking for them either.

The only drive really missing here that might have passed are the Seagate 600 Pro models. Those haven't been shipping long enough for me to recommend them yet.

Samsung's 840 models don't make any claims for power loss shutdown, and even the 840 Pro models fail this sort of testing. You have to step up to their SM843 to get something appropriate for serious deployments. The main problem with those drives is price/availability. If I need another Intel DC S3500, those are relatively cheap and plentiful, and I can even find them in my local Microcenter. Samsung's enterprise drives are much harder to obtain, which makes people nervous about deploying them. Nobody wants to return to the day when you needed rare and magical blessed drives in "enterprise" storage.

Manufacturers require 92 or better for certain high compression engines. This trend is increasing as manufacturers turn to more compression and forced induction to achieve CAFE fleet averages.

"Enterprise" grade drives are often faster, having better processors and more cache, and they don't do dumbass things like park heads every 8 seconds because the drive manufacturers have to listen to server and storage array manufacturers and meet their requirements to get certification for use in advanced storage systems.

You're an idiot. Please, stay away from any important systems. Just spend your time poasting on slashdot so you don't do any (more) damage.

The brand doesn't really matter. Storage Review has a user-submitted hard drive reliability database. Unfortunately you have to submit a HDD report to gain access to it, and the site's popularity has waned resulting in most modern drives being listed as insufficient sample size. But the older drives showed enormous variance within a brand name's lineup. e.g. The IBM Deskstar 75GXP (aka Deathstar) drive model had one of the highest failure rates in the survey. But the model which succeeded it had one of the lowest failure rates in the survey.

In other words, the model of the drive matters more than the brand.

RAID is dead Thunderbolt reigns supreme. You heard it here first folks.

In other news, various pieces of hardware simultaneously vanished from the earth as, given that "RAID is dead" and that "Thunderbolt reigns supreme", it was logically impossible for a Thunderbolt-attached RAID device to exist.

That's not what's on offer here, you don't get to decide HL2 should be in SSD, the caching algo decides which files in HL2 are cached based on how frequently you use it.

Yes, I know. I'm relying on extrapolating from anandtech's statement that game loading speeds are helped a great deal.

Likewise Windows caching algo decides the same for RAM. In WD's case, it works with that Windows caching algo, so they're one and the same algo.

So the WD's algorithm is perhaps not as good as Intel's Smart Response. Benchmarks would tell. That doesn't torpedo the idea of an SSD or hybrid drive, however.

Specifically you'd have to cycle through data heavy games that use MORE than 48GB in total (to ensure the 24GB of ram equivalent would be completely flushed)

That's unlikely. Typical buffer cache replacement algorithms are approximating LRU. If it were perfect LRU, then cycling 24GB of data plus 1 page would be sufficient. Since it's only approximate it would be rather less, but there are also "inefficiencies" in that figure that would raise it. In any case, you would be highly unlikely to require anything remotely close to 48 GB.

AND don't share common files, so for example the games don't reuse DX and drivers because they'd be in cache and would break up your long reads algo

You think it's DX drivers that cause game load times to be long?! No, it's level data and textures and models and such.

Flash is very slow to write

Write speed for HDD: "The range can be anywhere from 50 â" 120MB / s"
Write speed for SSD: "Generally above 200 MB/s and up to 500 MB/s for cutting edge drives"
source

Presumably WD stick some RAM on there too to cache the writes to flash.

No, they don't. They use better flash and they bank them.

If 4TB is the biggest drive you can get today, wouldn't densities have to increase by 15x to get to 60TB drives by 2016, not just "more than double"

Probably. The first 3tb was released June 2010. 4tb came out Oct 2011. Not exactly amazing growth, over a year for 1tb, at this rate we'll be 9tb in 2016. At this rate we will not see 60tb by 2016, and I say "we" meaning end consumer, maybe some lab monkey will see an areal density equivalent to 60tb, but it won't be available for sale. And for anyone wondering the answer is yes, the 4tb drives already use five platters, 800gb each, so they can't shove more platters in there to double capacity currently, they have to significantly increase areal density.

But while storage continues to increase, the types of media we store is not increasing in size. HD Video is probably the most common space hog on any personal computer at 25 to 40 mbit per second of video, about 11 to 18gb per hour, but once we have hundreds of terabytes what do we need more space for? For higher high definition video? At some point even video quality will surpass what the human eye can distinguish, especially from across the room.

And once we have hundreds of terabytes how do we fill the drive? Most of the content on my PC is downloaded, but internet speeds have not increased drastically over the years, I'm still at the same speed now as I was in 2000 and paying about the same amount.

They're putting the cart before the horse, they're offering us storage for something we don't have to store and that we can't even obtain through current technology.

Also... 1800gbits/sqin by 2016?

That 4TB hitachi drive is 446gbit/sqin http://www.storagereview.com/hitachi_ultrastar_7k4000_4tb_enterprise_hard_drive_announced

So if that increases to 1800.. we'll have 16TB drives NOT 60TB.

They may be slower than SSDs, but not by much

That's horribly incorrect. I liked the sound of hybrid drives as well when I saw the price... The 500GB laptop hard drives with 4GB Flash for $150, should be awesome... But I, not being an idiot, did some research, and sure enough, the reviews say it's not remotely comparable to a real SSD.

eg. http://www.storagereview.com/seagate_momentus_xt_review

It's faster than a drive without such a cache, and it might be a good option for a laptop, but even there I'd say a 32GB SD card would be cheaper, and will work wonders on FreeBSD with ZFS configured for L2ARC...

I have no particular interest in what anyone buys, but the comparison to real SSDs is a massively dishonest.

10/17/2011 : After months of end user complaints, SandForce has finally duplicated, verified and provided a fix for the infamous BSOD/disconnect issue that affected SF-2200 based SSDs.

wow, that's not something anyone wants to see, a bug in their hard drive. CPU I can replace, ram I can replace... pretty much everything I can swap out, but my hard drive is where everything is stored, I can't risk losing data because of a bug.

Intel just had problems too that cause loss of data:
"JULY 13TH, 2011 : Intel has recently acknowledged issues with the new SSD 320 series, where by repetitively power cycling the drives, some may become unresponive or report an 8MB drive capacity."

I was waiting on an SSD until they worked out the bugs and there were no articles about problems for awhile but with stories like these I'll keep waiting, it's just not worth the risk.

. . . a Storage Review experiment from over a year ago:

http://www.storagereview.com/western_digital_velociraptors_raid_ssd_alternative

They put WD Raptors in RAID 0 to form a high performance (yet still affordable) platter drive setup, and then faced them off against Western Digital's new (at the time, first) SSD. Makes sense, right? Except that WD's first SSD was a complete joke, an underperforming, laughably expensive POS that I forgot about a couple days after Anand's review. When I first read about it I couldn't help but think that WD was deliberately setting it up to fail. It was at the bottom of every benchmark yet priced higher than any other (MLC) SSD. They even put a jmicron controller in it for fuck's sake (not the infamous original one, but still . . .)! Storage Review's calling it a "mid-range" SSD is very generous at best.

Even so, this supposedly screaming platter drive setup could only occasionally hang with the bottom of the barrell of SSDs, and mostly lagged behind it. And as I said, this was over a year ago. It goes without saying that they didn't worry much about heat, noise, reliability (of RAID 0), or power consumption.

Anand doesn't even list platter drives in his benchmark results anymore because they'd skew the charts so badly.

As a previous poster said, a winning strategy is to get a SSD boot drive just big enough for your OS and programs, and use platter drives for everything else. And since the SSD takes care of your performance needs, you can get the cheapest, slowest, coolest, quietest platter drives. There are some cases where both high performance and high capacity are needed at once (like video editing) but they're not the norm.

"The random read performance depends on four factors, not one: the areal density, the seek/settle time, the rotational latency, and whether the data is in the drive's buffer cache already."

About 99% of random read performance depends on RPM, the other 1% is everything else. As for areal density and buffer: Techwarelabs short stroked a 7200rpm 1.5tb to 300gb and only got 10.3ms while 2004's (ok 7 yrs not 10 like I originally said) Fujitsu MAU3147 15,000rpm SCSI drive has a average random access time of 5.7 milliseconds. That Fujitsu is a 73gb drive with a much smaller buffer (8 vs 32mb) and much lower areal density than the 1.5tb drive and it's still nearly twice as fast at accessing data randomly. In case you think that's some kind of fluke Tomshardware tested a 450gb 15,000rpm SAS drive. It got 6.0ms.

I've tested 15,000rpm drives on xp, vista and 7. Boot-up and starting programs is night and day difference, but you don't really notice the difference any other time.

"it's fairly sequential reading (of multiple mostly sequential stripes in alternation), which is why the modern 5400 RPM drives spank older 7200 RPM and 10k RPM drives."

True, but very few people do a lot of sequential reading, unless you're doing a lot of video editing with large-ish files, ~100+ mB, where a older 50mB/sec 7200rpm drive would be beat by a modern 5400rpm 100+ mB/sec drive due to the areal density, but you have to go back several years to find a 7200rpm drive that only does 50mB/sec.

"I'm sure a modern 3TB drive would match or beat a 10 year old 15k RPM drive if you used short stroking* [tomshardware.com] on the modern drive."

You would be wrong. 2004's (ok 7 yrs) Fujitsu MAU3147 15,000rpm SCSI drive has a average random access time of 5.7 milliseconds. Tomhardware short-stroked a 7200rpm 250gb SATA drive down to only 12gb and only got 8.5 milliseconds. They also tested a 15,000 SAS in the same test and got 6.0ms.

Techwarelabs short stroked a 7200rpm 1.5tb to 300gb and only got 10.3ms

There is no replacement for RPM.

I don't know if they're a garbage brand, but OP is certainly right. You can read OCZ's own announcement, or read Anandtech's analysis in their Vertex 3 review. Storagereview did a comparison of the 32nm and 24nm Vertex 2's which is also worth a read. .

"Most video cards cap out at 2560x1600 per output, hard drives can be put in a RAID and solid state drives are damn fast, but the file sizes for videos at that resolution are absurd."

And 15 years ago a video card capable of 1920x1080 didn't even exist to consumers. In fact in 1996 the first consumer 3D gaming solution was released with an amazing 4 megabytes of RAM!

A good rule to follow is to never compare the hardware you're using now to what you'll be using 10 years from now. I have no doubt we'll have CPUs with hundreds of cores and SSDs (or whatever they're called by then) that are several terabytes and transfer data at gigabytes per second. Remember, largest hard drive in 2001 was 100gb for $300 which could be read at an amazing 26 megabyte/second!

Like reminiscing about old video cards? Check out the history of 3D Graphics

DISKEEPER 2011 & it's "HYPERFAST" feature for SSDs:

http://www.storagereview.com/diskeeper_2011_now_available_includes_ssd_optimizer

Pay attention to "HyperFast" + "Intelliwrite" featureset...

(As it may be the "cure" here for that which you speak of (it sure sounds it @ least), which folks that use SSD's for browser cache hoseups are 'victim to'... )

Incidentally, as an "addendum" as well?

PerfectDisk by Raxco has a "background defrag" perfect write feature also, but, not 110% SURE it's "geared to SSD optimizations" only though...

APK

P.S.=> That's my "tit-for-tat" info. for "U", in return for your tips on browser cache hassles on random writes of smallish files on FLASH SSD's... hope it helps, or, is at least interesting to you & other FLASH BASED SSD users that have experienced the write issue you speak of! apk

Actually USB mass storage can support >512 byte sectors easily

E.g.

http://forums.storagereview.com/index.php/topic/28840-so-i-bought-the-new-3tb-goflex-desk/

This actually presents a 4096 byte physical sector size to the OS. Handily this means that MBR partitioning will still work with it - that has a limit of 2^32 sectors but with a 4K sector size that is 16TB..

Windows is fine with >512 byte sectors at least post boot but I don't think it's possible to use them over ATA on any current OS. I.e. there are Advanced Format drives with 4K sectors but currently they all emulate 512 byte sectors over ATA. Probably moving to 4K native sector size on ATA is going to take some time since that requires changes to the ATA spec, drives, Bios and OS boot code.

Since the USB mass storage driver has supported >512 byte sectors for ages because CDs and DVD Roms have 2K byte sectors it's actually easier to get >2TB drives working over USB.

The reason I know about this stuff? I wrote this free tool

http://www.ridgecrop.demon.co.uk/guiformat.htm

It so happens that it works fine with the 3tb Goflex USB drive because it has a 4K physical sector size. FAT32 has the same limit 2^32 sector limit as MBR.

So you can have 16TB FAT32 volumes to share your AVI files between your PC, Mac, Linux box and games console/media player. 16TB is a lot of movies, maybe even all of them.

Now some might say that it's insane to use such an old format. That's sort of true but FAT32 is good because it's so widely implemented. It's hard to imagine anything else being supported across such a wide range of devices. Also if all you're doing is streaming AVI files FAT32 is actually good enough - because it is so simple it is trivial to implement it and get stability and performance very close to the raw performance of the device. More modern file systems are not like that.

Depends on what kind of a secure delete you're referring to. Simply overwriting a file -- ie. the way to securely delete it on a normal hard disk -- won't accomplish much: the meaningless data that's meant to overwrite will end up somewhere else, since there is no easy way to tell the controller where to actually put stuff. So, yeah 90% might be left.

However, if you securely wipe an entire drive, by writing $capacity bytes to it, nearly everything on the drive will be irrevocably gone. I say nearly because all SSDs use overprovisioning to a certain degree: there is more physical flash memory available than the capacity reported by the controller. The excess space is used to increase performance and reliability. It's conceivable that a file gets written to a block that's never again written to because it's permanently replaced by a block in the overprovisioned area. AFAIK, regular HDDs also have a little bit of extra capacity to improve reliability, but it's nowhere near as much: the new 25nm OCZ drives has 128 GB "raw" capacity of which only 118 GB can be used. (Slightly more, 120 GB, can be used in the older 34nm version. Source: StorageReview.)

Of course, none of this takes into account what I assume TFA is talking about: that the controllers on SSDs will often permanently wipe blocks without the user going to any great lengths (beyond, you know, unlinking the file), in the normal course of operations that are, in fact, required for the enduring performance of the SSD.

http://forums.storagereview.com/index.php/topic/28840-so-i-bought-the-new-3tb-goflex-desk/page__p__262945__hl__seagate%203tb%20goflex__fromsearch__1&#entry262945

You definitely need a 64 bit OS to fully see the drive. Well unless you can manage a hack like using larger block sizes. You also need a controller that supports 3TB and also has Linux drivers (not that hard to find).

For instant-on, you could have the computer boot in a completely clean state then freeze that state to file. I practically guarantee that unthawing that state, then tweaking it afterwards (kill -HUP is your friend) will be faster than any staged booting or threaded booting could ever be.

It takes Lucid 23 seconds from power-on to a usable desktop on my HP Mini 110. Its WD Scorpio 160GB hard drive (model WD1600BEVS) benchmarks at 49MB/s sequential read, assuming large contiguous chunks.

What you're suggesting is basically restoring from hibernation each time, which can be done today. On my hardware, if the restoration process has no overhead above just streaming the hibernation file to RAM (including delays that the cold boot has to deal with, like BIOS testing), and the state file is completely defragmented (as if it had been saved to a pristine swap partition), it can be faster than a cold boot only if I've saved less than 1.1GB of state. If I'm saving more state than that, or if it takes time to bring the network up, or any hardware has changed and has to be redetected, or [...], then it's actually faster just to reboot the whole thing.

It's hard to find...
http://www.storagereview.com/western_digital_velociraptors_raid_ssd_alternative
http://www.pcper.com/article.php?aid=878&type=expert&pid=8 (this website uses iometer on all the ssds they test i think so can just look at ssd reviews like this one to compare numbers)

I have citations for everything you asked for, and then a few more:

Wear-leveling is done for bad sectors on a HDD, not as standard practice.

1) I still was considering that wear-leveling. Just after-the-fact wear leveling. :-)
2) However, Western Digital drives do have "Preemptive wear leveling" as a standard practice.

Wrong. I just ran a benchmark and...

On a HDD, writes are followed by an immediate read to verify the data wrote correctly. Take a look at some of the benchmarks on storagereview.com Compare the random writes to random reads: the writes are always slower. StorageReview doesn't compare continuous writes, because that is rare. However, you will find that there are drives specially designed for continuous writes for A/V purposes, that are made to address this issue.

Try to defrag a MLC drive a few times and it'll be dead in a week

My turn. [citation needed]. I'll provide my own references.
According to Kingston

For USB Flash drives, Toshiba calculated that a 10,000 write cycle endurance would enable
customers to “completely write and erase the entire contents once per day for 27 years well beyond the life of the hardware.”

Intel was more conservative with their report

...the X-18/25 SSDs have a mean time before failure (MTBF) rating of 1.2 million hours, which is on par with modern server hard drives. In addition, he claimed that the drives can withstand a workload of 100 GB worth of writes a day for five years.

They then go on to explain how it was actually hard to even write 100GB per day.

not degradation of the media itself.

Magnetic media does degrade. For one thing, it warps over time. I can't state all of the methods of failure.

Look, SSDs are great... But they are not ready to replace HDDs.

I agree. Just not for some of the reasons the Wikipedia article lists.

And SCSI. And FireWire.

Apple invented SCSI? News to me. FireWire, yes, but SCSI, no. They may have adopted it to a greater extent than other personal computer makers, but they didn't invent it.

According to this article, IBM switched over to glass substrate in consumer hard drives circa 2000. The surface of the platter is smoother so R/W head flight height can be lowered to accommodate higher data density.

Which is why I thought the consumer-level advice "hammer your hard drives" may be a bad idea if Joe Bag'o'donuts uncases the platters first. Can you say lawsuit for glass splinter injuries?

If you take a look at the newegg reviews, you'll find 16% of them give the 1TB 7200.11 drive a 1 star review, most of which are because of DOA or D shortly after A. So it's not just you who noticed.

Seagate's Barracuda line had a good run with high reliability for quite a while. If you check the reliability database at storagereview (unfortunately you have to go through some trouble to become a member and see the data), the Barracuda ATA III, IV, and V are ranked near the top--92, 90, and 96th percentile respectively. Then things went way downhile--7200.7 hits 88, the 7200.8 at 49, and the 7200.9 at 43. That matches my own anecdotal experience.

Sometime after the 750GB drives came out reliability took a further dive south. I believe that was caused by switching a large amount of production to a new plant in Thailand (the reliable models came out of Singapore). That seems to be the inevitable way hard drive manufacturing works--whenever some company moves to a new facility, quality dives for a few years afterward. I predict that 5 or 10 years from now talk will be about how reliable the old Thai drives were compared to the new junk coming out of [new country of origin].

There is a good answer (as well as a lot of more information and history about hard disks) here:

http://www.storagereview.com/guide2000/ref/hdd/op/actMultiple.html

That's a pretty fundamental part of information theory - communication in a noisy channel. If your communications (or data storage) are digital, you can overcome any level of random noise (error) at the cost of degraded transmission rate (increased storage requirement). Before CDs, it was (and still is) most prevalent in modem protocols and hard drives. Modern hard drives would probably be impossible without it - read errors are the norm, not the exception. It's just hidden from the high-level software by multiple levels of error correction in the low-level firmware.

mirroring should speed up read performance relative to a single drive

No, not unless you're doing multiple concurrent IO. It's not RAID-0, it can't interlace IO's to get greater sequential transfer rate. In fact if it's done naively you get reduced STR by bouncing the IO's between the disks.

My 1000G Seagates do around 109MB/s, though they do drop around 60MB/s on the inner edge. See StorageReview; the ES.2 is basically the same as the 7200.11, and they have the SAS version topping out around 120MB/s.

Good Source is Storage Review
http://www.storagereview.com/php/benchmark/bench_sort.php

The top 34 drives all do at least 54mb/sec MINIMUM and at least ~80MB/sec maximum. The top 15kRPM cheetah doing 82.7-135MB/sec.

If i were to pull a number out of my ass I would say 78-135MB/sec (min/max) on the new 1.5TB drives.

I would say if you have 750gig seagates and you are only getting 25MB/sec you have a bottleneck. Those drives should do a MINIMUM of at least 40MB/sec...

Server drives with high density need to be faster (seek and transfer times) to support more multiple users accessing different sequences of the disk's storage addresses in rapid interleaved succession.

But personal drives don't need as high speeds for one person's use, especially when the high capacity is for large media content objects that are stored unfragmented. We don't need to spend the money on transfer speeds so much faster than our playback speeds that it's never used. Large builtin caches are useful for real random-access data in small chunks, like programs or numerical datasets, not media.

Blu-Ray's max transfer speed is 54Mbps, though that's for recording - 48Mbps is max playback. 3x for buffering during FWD/REV scanning playback would be 144Mbps, 2.25MBps. Big drives currently recommended for personal use, like Seagate's 1TB Barracuda ES.2, get at least 53MBps transfer, over 23x as fast as the fastest it will ever really be asked to deliver. If it weren't so unnecessarily fast, maybe it would cost less, and an array of them for the same hundreds of dollars would hold more content.

With 50GB Blu-Ray HD titles to store, getting more sets of 20 titles in each HD in a RAID is a lot more important than getting them faster than they can be played.

These ones look pretty awesome; they're quieter than pretty much every other drive (including a bunch of laptop drives), bar only a Samsung Spinpoint and WD's low RPM GreenPower drives (which are great if you're not bothered by slow seeks, btw; the first drives I'm happy to passively cool in about 8 years).

Very tempting.

The 300GB Seagate Cheetah 15K.5 is $675.00 at Dell (source: Google, while the Raptor is (supposedly) about $300.

That's 2.25 times the cost per megabyte.

According to this performance database (choose IOMeter 8 I/O. I can't link to it directly, it doesn't seem to support that), the Seagate drive does 293 IOPS vs. the Raptor 3000's 228, so it's only 28% faster (on an 8-deep workload, which is a fairly common one, maybe a little deep).

Cost-per-IOPS wise, the Raptor blows the 15K SCSI drive away. Of course, the Seagate is an SAS drive, which is far more robust for large server installations and such, but for smaller ones, I think the Raptor would be fine. Of course, for most smaller ones, a cheap-o 250GB 7200 RPM RAID1 array would be fine, too.

Power usage is 60% lower than any drive ever encountered (see earlier post). Apparently the huge heatsink is epoxied (or something like that) onto the drive. Not very bright on WD's part, unless I'm missing something.

The enterprise version is supposed to use a standard connector, so those who want their laptop disk IO to outperform most desktops, including most RAID0 arrays, may be able to use those.
For reliability, I have an old 74GB Raptor that's still working fine, but StorageReview's reliability benchmark says they are more reliable than "12%" of other drives. Not that it's scientific, but it isn't promising.

Power usage = heat.

From the StorageReview.com article:

When spinning up from a cold start, the WD3000BLFS maintains its prowess with a very economical showing on its 12V rail. At just 9 watts, the VelociRaptor weighs in a full 6 watts (66%!) lower than any other drive SR has ever encountered.

I think the heatsink is mostly for show, and to make the drive fit into a normal case. Still, it would be nice if they made it easily removable.

The review is up on on StorageReview.com . You can use the database to compare this drive to every other drive out there in different kinds of tasks.

The review is up on on StorageReview.com . You can use the database to compare this drive to every other drive out there in different kinds of tasks.

...is the StorageReview.com Hard Drive Reliability Survey: http://www.storagereview.com/map/lm.cgi/survey_login

You basically input all the hard drives you possess into their database and then they let you see the statistics collected so far.

When one of your drives fail, you ought to update its status (at what age has it failed).

The database still contains a bit sparse information, but it's still the best I could locate on the Internet.

To make this sort of test work, it must be run over a much longer period of time. But in the process of designing, building, testing and refining disk drive hardware and firmware (software), there isn't that much extra time to test drive failure rates. Want to wait an extra 9 months before releasing that new drive, to get accurate MTBF numbers? Didn't think so. How many different disk controllers do they use in the MTBF tests, to approximate different real-world behaviors? Probably not that many.

Could they run longer tests, and revise MTBF numbers after the initial release of a drive? Sure, and many of them do, but that revised MTBF would almost always be lower, making it harder to sell the drives. On the other hand, newer drives are certainly available every quarter, so it may not be a bad idea to lower the apparent value of older drive models.

So, it's better to assume a drive will fail before you're done using it. They're mechanical devices with high-speed moving parts, very narrow tolerable ranges of operation (that drive head has to be far enough away from the platters not to hit them, but close enough to read smaller and smaller areas of data). Anyone who's worked in a data center, or even a small server room, knows that drives fail. When I've had around two hundred drives, of varying ages, sizes and manufacturers, in a data center, I observed a failure rate of five to ten drives per year. This is well below the MTBF for enterprise disk array drives (SCSI, FC, SAS, whatever), but drives fail. That's why we have RAID. Storage Review has a good overview of how to interpret MTBF values from drive manufactures.