Terabyte Hard Drive Put To the Test

EconolineCrush writes "As a technical milestone, Hitachi's Deskstar 7K1000 hard drive is undeniably impressive. The drive is the first to pack a trillion bytes into a standard 3.5" form factor, and while some may argue the merits of tebi versus tera, that's still an astounding accomplishment. Hitachi also outfitted the drive with 32MB of cache—double what you get with standard desktop drives—making this latest Deskstar a leader in both cache size and total capacity. That looks like a great formula for success on paper, but how does it pan out in the real world? The Tech Report has tested the 7K1000's performance, noise levels, and power consumption against 18 other drives to find out, with surprising results."

Comments

The author has some problems with his arithmetic

[Don Sample]

He spends a lot of time talking about the difference between binary and SI terabytes and gigabytes, and then comes out with:

Back in the day, the gap between decimal and binary capacity wasn't big enough to ruffle feathers. Gigabyte drives were only "missing" 24 bytes, and that was easy to swallow.

Um, 24 bytes is the difference between kilo meaning 1000 and kilo meaning 1024. A binary gigabyte is 1,073,741,824, or 73 megabytes bigger than an SI gigabyte.

Conclusion in the article:

[Futurepower(R)]

Conclusion in the article: Too expensive.

They may have meant 24 k

[FoamingToad]

But they'd have still been way off.

For a decimal megabyte versus a binary one, there's 48 1/2 KB difference.

For a gigabyte, there's about 70 megabytes difference.

The only case where you'd only lose 24 bytes would be if you had a kilobyte drive.

F_T

O RLY?

[_Shorty-dammit]

http://stason.org/TULARC/pc/hard-drives-hdd/ibm/WD A-L42S-40MB-3-5-HH-IDE-AT.html Hard Drive: IBM: WDA-L42S 40MB 3.5"/HH IDE / AT Cylinders: 1067 Heads: 2 Sectors per track: 39 Bytes per sector: 512 1067 * 2 * 39 * 512 = 42,611,712 bytes 42,611,712 / 1024 = 41613 kilobytes 41613 kilobytes = just over 40.6 megabytes This was sold as a 40MB drive. Not a 41MB, 42MB, or 43MB drive. A 40MB drive. And that's just what it was, a 40MB drive. So, I'm sorry to tell you, but lying about the drive's size was *NOT* always the way it was. This is just one example. And, no, I don't care for finding out exactly when manufacturers started lying about capacity. They did, and that's enough for me.

Re:Data loss

[blackicye]

BTW: Turn off S.M.A.R.T. This is like the indication of an ink cartridge: When the maker thinks you need a new
drive.

In my experience, when S.M.A.R.T. tells you a drive is dead or dying, its not kidding.

The history (as I remember it...)

[DusterBar]

The "clever" marketing company was Atari with the 520ST - they wanted to make it sound better than the Amiga with 520K of memory (it had 512K like anything else, but it was 520 in marketing terms). The same reason they has the 1040ST.

Note that it was sometime after that point in time (don't have the exact year) that some hard drive manufacturers started to play the same games. (Only with megabytes). Back then it was common to look at a 30meg vs 32meg drive and pick the 32meg drive. So when a marketing person figured out that a "real" 40meg drive could be called 42meg "unformatted" and get away with it, well, they did. And the other manufacturers followed and, well, everything was different by the time 1990 happened... (or so, maybe 1991 for the last holdout)

It really does not matter much now - but when different manufacturers followed different rules, it was a real problem.

(ps - Jack was always pushing the marketing envelope - albeit I can not claim to know if he did come up with the 520-vs-512 idea himself, he did push it rather hard)

Re:Data loss

[Ex-MislTech]

Some problems with RAID 1+0:

Not all hardware controllers will allow you to do a reconstruct to add more
space and extend the partitions later on RAID 10 or 1+0.

Recovering from a failed 1+0 is ok if it is a "simple" failure.

I have had better luck recovering RAID5's than 10's or 1+0's.

Re:Data loss

[Eivind]

That's nonsense. It isn't even true in theory. (at some point the remaining charge is below the noise-floor) If it wasn't you could store an infinite amount of data on a drive by simply filling it once with dataset1, overwrite by dataset2, overwrite by dataset3 and so on. You claim dataset1 will always be recoverable, so in this method, you could recover each of the sets and have stored triple amounts of data on the drive. You claim *any* amount of overwriting will be insufficient, so I guess I can store 1000 datasets on the drive then. Cool. Hint: The real world doesn't work like that

Secondly, even if in theory you where rigth (which you aren't), in *practice* most data is not valuable enough that theres much real risk that anyone will recover it, even after something as simple as a one-time-all-nulls overwrite. (which is just about the suckiest overwrite you can do) Yes, in that case an expert lab *can* recover it, but odds are it won't happen.

In practice, if you do the standard wipe, which is usually some variant of all-nulls, all ones, 3 times random, there is -zip- chance that anyone will be able to get at the data that was once on the platter.

Now, what many (clueless people) do are "format" the drive or "delete" the files. These functions don't overwrite even once 99% of the platter, so files removed in this manner are certainly recoverable -- they're there in plaintext, just not referenced from the filesystem anymore. Something as simple as "cat /dev/hda | strings" will recover huge amounts of text from a hard-drive which has been erased in this manner.

Power-of-10 prefixes are the norm in IT

[this great guy]

Contrary to common belief, power-of-10 prefixes (as used by disk manufacturers) are much more commonly used than power-of-2 prefixes in the IT world. People claiming the contrary are wrong. Here are a few examples:

• A 128 kbit/s audio stream is 128 * 10^3 bit/s (power of 10)
• A 100 Mbit/s ethernet card is 100 * 10^6 bit/s (power of 10)
• A 480 Mbit/s USB2 link is 480 * 10^6 bit/s (power of 10)
• A 500 GByte disk is 500 * 10^9 bytes (power of 10)
• A 56 kbaud modem is 56 * 10^3 baud/s (power of 10)
• A 1.5 GHz processor is 1.5 * 10^9 Hz (power of 10)
• A 6 Mbit/s DSL line is 6 * 10^6 bit/s (power of 10)
• A 650 MByte CD is 650 * 10^6 bytes (power of 10)

It is a total mystery to me why people think that power-of-2 prefixes should be the norm, when the only few places where they are used are to refer to the size of files and RAM sticks.

Spread the truth. Mod me informative ;-)

Re:Power-of-10 prefixes are the norm in IT

[this great guy]

So you are still not convinced ? Here are some examples not based on bits or Hz:

• A 1x 250MB/s PCI-e lane is 250 * 10^6 byte/s (power of 10)
• A PC3200 DDR400 memory stick is 3200 * 10^6 byte/s (power of 10)
• A 56 kbaud modem is 56 * 10^3 baud/s (power of 10)
• A 650 MByte CD is 650 * 10^6 bytes (power of 10)
• A 300 MB/s SATA link is 300 * 10^6 byte/s (power of 10)
• A 4000 MB/s HT1 link is 4000 * 10^6 byte/s (power of 10)
• And of course, a 500 GByte disk is 500 * 10^9 bytes (power of 10)

Re:Visit our site!

[LarsG]

I've RTFA, and still don't get what the 'surprising results' is supposed to be.

It has huge capacity - check.
It is noisy and sucks power - check.
It is not a speed champion - check.

Not exactly surprising for the first 1TB drive on the market.

Re:RAID 6 Please

[Anonymous Coward]

Because it isn't safer, at least not unless you have a battery-backed interface card *anyway*. And usually the only cards with battery-backing are... hardware raid cards. I have been known to use linux-md on an old hardware raid card, when the card had a very slow processor. But these days, raid cards have processors in the 300-600 MHz range, AND ASIC or FPGAs to do parity. They're more than adequate.

Also, linux-md doesn't guarantee ordering, which hardware-raid cards, as they're intended for use with oracle and friends, do.

Re:Data loss

[zeromemory]

Not all hardware controllers will allow you to do a reconstruct to add more
space and extend the partitions later on RAID 10 or 1+0. Likewise, many hardware controllers won't let you extend a RAID-5 array, (unless they implement some dynamic stripe size hack, a la ZFS's RAID-Z).

Recovering from a failed 1+0 is ok if it is a "simple" failure. Please explain what a !simple failure would be. Here, let me give you a 'simple' failure case where RAID-5 would be pretty difficult to recover from: a drive fails in your RAID-5 array, and you lose power or experience another hardware failure shortly afterwards, before you can replace the drive. Whoops, you just became another victim of the RAID-5 write-hole (see the section under RAID-5 performance).

OK, here's why we use RAID-10 at my installation: it provides great performance and can survive multiple drive failures without the overhead of something like RAID-6. RAID-10 also has no 'write-hole'. Don't just take my word for it, though, check out this article from Adaptec comparing the merits of all the basic RAID levels and their nested brethren.

Re:Data loss

[Gr8Apes]

I think you've missed the point - RAID 1+0 allows you to fail up to n drives (with 2 n needed to build the array).

Additionally, should a drive fail, rebuilding will only marginally affect your performance, degrading it by a fraction compared to a RAID 5/6 rebuild. (Only 1 drive is affected out of your stripe set, the rest perform at peak operational speed)

Re:Data loss

[encoderer]

Yesterday at a Best Buy in Ohio a guy and his wife were looking at 42" Sony LCDs. There was a 1080P for $1899 and a 1080i for $1599. Guy asked the sales associate what the difference was.

Sales associate, I shit you not, said "The "P" is actually a newer product. It is 7 minor revisions later. We still carry the "i" because it's still very popular. The same thing happens with our wireless equipment, too. the "N" version is out, but most users are still buying the "G" Version"

I approached the guy after the sales associate left and said "listen, that guy has no clue what he's talking about. I is interlaced, P is progressive. On an "i" it's drawing 540 lines every frame, on a "p" it's drawing all 1080. Go with the "P" if you can afford the difference. It's worth it"

Re:Data loss

[walt-sjc]

I lost 3 drives out of 6 within a few hours of each other. Raid 1+0 saved my bacon. Zero down time. Got the email alert about the first drive, and scheduled a trip to the datacenter. Then I got the other two back to back a couple hours later. These were all 15K rpm SCSI drives which had survived a 2 week stress test burn-in, and had been in production for about a year, so it was totally unexpected. In another case, I lost 2 drives in a Raid 5 and had to resort to restoring the machine from backups - a day lost. Raid 6 performance is even worse than Raid 5, so I personally see no point - YMMV. Raid 5 and 6 rebuild time is also VERY slow compared to 1+0, taking 3 times longer in my testing.

Anyway, what's that old saying? Expect the unexpected? When you buy a pile of drives, you are likely to get the batch from the same manufacturing line, day, etc. This probably also increases the chances of simultaneous failures if there is a physical quality problem. If you have two fail, expect a third. I generally don't mix up batches because I want to know where all the drives from a particular batch are, but maybe I should.

Re:Ok... but 992, 977, 1023, 1011, 973 or 1005?

[poot_rootbeer]

Maybe 1023 tons of TNT is what fits on a standard truck, so it would be handier than that stupid 1000 for a kiloton.

Are those "long" tons (2240lb), "short" tons (2000lb), or "metric" tons (1000kg)?

Ambiguous terms of measurement do exist outside of the computer industry, too -- which, I should point out, is actually "the software development industry" plus "the hardware manufacturing industry" plus "the IT service industry" and so forth.

Drive manufacturers have always used base-10 prefixes to describe the capacity of winchester drives. It's not a marketing ploy, it's historic convention.

Re:Data loss

[operagost]

RAID-6 may seem inefficient, but it's superior to RAID 10 because it is capable of recovering from two drive failures, whereas RAID 10 may recover from two drive failures. It is possible to lose a second member of one of the mirrors before the first is rebuilt. In RAID 6, you already have enough parity data online to recover immediately from two simultaneous failures.

Re:Data loss

[Feanturi]

What are people doing with drives to make them fail?

I've got the same question, as I've gone through a lot of hard drives over the years but only due to upgrading, not failure. The only exception was the IBM Deskstar GXP75 that had the whole click of death thing going on. I don't count that one since it was a known issue that resulted in a class action suit, which I didn't bother to take part in. The first one failed within a month, so I replaced it at the store, and the replacement failed after a day. Replaced again. The third one failed after a week but I was tired of going back to the store by then so tried an experiment - the click of death was kicking in somewhere near 500MB after the beginning of the drive so I repartitioned it to leave the first 500MB unpartitioned. My experience with the drive up to that point told me that wherever the click of death manifested, it would consistantly happen in whatever part of the drive it first happened at. That drive has been in constant use ever since then (it's been like 5 years or so by now hasn't it?) and still works great, since it never accesses the 'bad part' anymore.

Re:Data loss

[Mr Z]

Right, but if the wrong 2 drives fail, you're hosed. You can lose up to N drives (out of a 2*N array) only if you lose 1 drive from each mirror pair. It's possible to hose a RAID 10 array of any size by losing just two drives if they're the two drives that mirror each other.

Re:Data loss

[afidel]

The best is something like what Xiotech implements, we do virtual RAID10 across all our spindles and it makes sure that no data and redundancy block are in the same drive bay. That way you can lose both power supplies or fibre controllers in a disk cabinet and not lose any data. It can do the same with RAID5 but performance is limited by the speed that the controllers can do the parity calculations whereas with RAID10 we are limited only by FC bandwidth or the number of concurrent I/O's we are trying to push (we've tested to 18,000).

Re:tebi? shut up. 1 terabyte drive still NOT here

[clarkcox3]

So? "Byte" is not an SI unit.

KB, MB, GB, TB, etc. have had a well-defined meaning for decades (probably over a half century by now). According to The Oxford Pocket Dictionary of Current English:

n. Comput. a unit of memory or data equal to 1,024 (2^10) bytes.

... so get over it, a kilobyte is 1024 bytes.