Slashdot Mirror
Seagate Announces First SSD, 2TB HDD
Lucas123 writes "Seagate CEO Bill Watkins said today that the company plans to put out its first solid state disk drive next year as well as a 2TB version of its Barracuda hard disk drive. Watkins also alluded to Seagate's inevitable move from spinning disk to solid state drives, but emphasized it will be years away, saying the storage market is driven by cost-per-gigabyte and though SSDs provide benefits such as power savings, they won't be in laptops in the next few years. A 128GB SSD costs $460, or $3.58 per gigabyte, compared to $60 for a 160GB hard drive, according to Krishna Chander, an analyst at iSuppli. 'It will take three to four years for SSDs to come to parity with hard drives,' on price and reliability."
Every news source has merged those two statements together, and every time, my brain gets stuck on it.
Seagate is announcing two seperate products. One is a SSD and the OTHER is a 2TB hdd.
Modding Trolls +1 inciteful since 1999
This
I can't help it. I remember buying my first two gig drive for $780, back when the dollar was worth squat. Now of course, the value of the dollar is rapidly approaching diddly-squat.
-jcr
The only title of honor that a tyrant can grant is "Enemy of the State."
The SSD from Seagate is a total "Me Too!" product. Seagate thinks they are in the "Mass storage" market, but they are not. They never have been. Their market is the one that includes "rotating magnetic platters". The only reason they are trying to break into this market (that they continuously decry as useless, futile, and too expensive) is because they are afraid of what "might" happen ten years down the road.
It's so nice to see a company that fought this at every step pretend to embrace it.
Modding Trolls +1 inciteful since 1999
Technically speaking, if it were always about price vs performance, we'd all be running last generation AMD's, using CD-R's and the like. In reality, you don't pay proportionally more for extra performance, you pay EXPONENTIALLY more.
/. readership, a 20% boost in performance is worth a 200% increase in price, especially considering how cheap computing equipment is these days, compared to the utility it offers.
For the average consumer, SSD's aren't yet the way to go, but for what I'd bet is a good proportion of the
which utilizes both SSD and a mechanical disk to get the best of both worlds in a way similar to processor caches L3 >> L2 >> L1. Ofcourse, current drives already use buffering but the buffer data gets lost when the drive is switched off.
They called me mad, and I called them mad, and damn them, they outvoted me. -Nathaniel Lee
I've told you a million times not to exaggerate...
Is it that fucking hard to include the cost per gigabyte of the current hard drives ($0.375/GB for the example given)? Why quote one $/GB figure if you can't be bothered to include the other?
[b.belong('us') for b in bases if b.owner() == 'you']
SSD will never reach parity with hard disks because of the economics of spinning disk storage. Yes folks, a 160GB drive costs $60.
SAMSUNG Spinpoint F1 HD103UJ 1TB 7200 RPM SATA 3.0Gb/s Hard Drive - OEM. Cache: 32MB. Form Factor: 3.5". $184.99
Seagate Barracuda 7200.11 ST31000340AS 1TB 7200 RPM SATA 3.0Gb/s Hard Drive - OEM. Cache: 32MB. Form Factor: 3.5". $209.99
Next year these will be 4TB, 8TB, 16TB? $100-$200 range. Call me on it; by December 2009 (i.e. in 2009, next year) it'll happen. Where will we see the SSD price point?
Support my political activism on Patreon.
I'd buy a $460 128 gigabyte SSD in a laptop. Not to long ago these options were about $1000. If you do this right (and often it's been done wrong) you get better performance, much longer battery life, and enhanced reliability. With the right software monitoring of repeated writes, you could also know about hard drive failures coming in advance. That's fantastic, in my book. $460 is still a tad high, but I'd bite.
I remember my long-former managers happily paying nearly $10k each, for the damned things...
Quo usque tandem abutere, Nimbus, patientia nostra?
Spinning hard disks will go the way of tubes in ten years, more likely faster than that. Scaling the manufacturing up will drive SSD drive costs down. There are long-life reasons why:
- SSDs aren't as vibration sensitive (both will not take a bullet, but only SSD can likely survive a normal drop of 2M on to concrete)
- SSDs don't have the temperature/altitude constraints
- SSDs don't have latency and no rise/shutdown time for green needs, in fact, they use hardly any power at all
- SSDs are generally faster, although there are algorithms needed in flash to prevent bucket overuse because reads are almost infinite, but writes are not
- SSDs take less in terms of precious metals and present fewer QA problems
- No electromechanicals to wear out.
The price point? Going down. It's an obvious solution to a long time problem. Magnetic versus flash storage will tend to favor flash, as magnetism decays sooner than flash will-- when flash is written to correctly.
---- Teach Peace. It's Cheaper Than War.
For two reasons. First and foremost, low power consumption. Secondly, we have already passed the sweet spot in the storage capacity needed for the applications most people run, particularly on laptops. Add to that the fact that current HD form factors are an extremely good fit for SSD units, and the writing is on the wall.
So what will happen is pretty obvious. Laptops are going to push SSD storage into the mainstream, giving it the critical mass needed to start the research bandwagon rolling, and 5-10 years after that happens hard drives will become the 'new' tape storage and most production systems will be using SSDs.
Even more pointedly, with power costs being the premium concern for data centers these days, and the hard drive being the only thing left in the computer that can't be engineered down to near 0 power consumption when idle (short of spinning it down, which has its own problems), my expectation is that large commercial concerns will see a huge cost benefit to using SSD storage despite the higher front-end cost of purchasing it.
-Matt
The false dichotomy is that it's an either/or thing.
There will be SSD components with high speed and low power and their price/GB will decrease very quickly. Largest capacities will always be expensive. For a long time they will cost more than magnetic media, but it probably won't always be so. Their speed and reliability will improve as vendors build out the drive intelligence that abstracts the physical media from the logical media and parallelize atomic access with internal RAID to compensate for the slowness of individual cells. These products will sell to users who are interested in their benefits and the manufacturers will make lots of money. Ultimately the speed of random reads and writes of SSD media will be limited only by the interface as solid state components are "always on" and each block of data is as accessible as any other.
Magnetic media will continue to drop in price as well. As storage increases today's hard drives will find their way to the recycling center in record time. The optimal price/performance will continue to improve as will maximum capacity. Speed will not increase as much, particularly with random reads and writes, because the data is still stored on a rotating physical object and a physical read/write head must move to the correct track and wait for the data block to fly under the head before data can be read or written. These products will continue to sell well for a long time to users interested in their benefits and the manufacturers will continue to make lots of money.
Both will be popular for a long time. There are other technologies in the works also.
Ultimately at 60MB/s it takes 1,000,000 seconds (11.5 days) to write 60tB to a (currently theoretical) rotating platter drive. At 6tB/second (interface TBA) it takes 10 seconds to write the same data to a solid state device. The ultimate winner in this one is clear, but it will be a long time.
Let me be the first to say: "that's a lot of porn."
Help stamp out iliturcy.
Hutz: All right gentleman. I will take your case. But I will require a thousand dollar retainer.
Bart: A thousand dollars. But your ad says "no money down".
Hutz: Oh, they got this all screwed up. [corrects ad with felt-marker]
Bart: So you don't work on a contingency basis?
Hutz: No, money down. Oops, I shouldn't have the Bar Association logo here either.
...paying $400+ for a 128gb SSD to replace the standard sata drive in my laptop as long as the performance was truly better and the battery life was that much better.
The problem with quotes on the internet, is that nobody bothers to check their veracity. -- Abraham Lincoln
It's not impossible to implement that functionality with a dumb SSD and HDD. The easy part is unionfs -- done. The hard part is determining with sufficient accuracy what files are unlikely to be written again -- a first cut could just consider some directories, MIME types and/or file extensions more or less likely to be rewritten than others. The ugly part: file metadata has to be present for both file sets at all times (or at least all directories which are split across both devices), metadata might be changed frequently, the HDD must be on for as little time as possible, and writing to flash must be avoided as much as possible. The only way to satisfy all those constraints is by reading and maintaining a complete write-back cache of the HDD's inodes and dirents in RAM at mount time, flushing dirty entries whenever the HDD spins up and writing through whenever the HDD is on. At 144 bytes apiece a cache for a typical homedir/archive disk could eat up a sizable chunk of RAM. While we're dreaming, database engines could even be optimized to read only from the SSD-portion of a hybrid drive if a particular data point had not been written to in over N minutes, or since the last collation (explained later), but would write to the platters, and then during quiet cycles, it could do a collation. The collation would move data which was on the platters, but which did not have a pattern of large volumes of writes back to the SSD volume. An equal amount of battery-powered RAM as cache and journal for a traditional HDD would under most real workloads beat RAM+SSD or HDD+SSD. If you really wanted to identify (manually or otherwise) cold tables and load them into flash SSDs, the database engine will probably still load and cache them in main memory anyway (costing all of a few extra milliseconds), and any RAM not used to cache those tables can be used to speed up temporary tables or for dynamic caching. (compare Amdahl's Law) And... I'd like a pony... NOT YOURS
The usual structure of a storage hierarchy is that each level contains a fast, small subset of the next level. A consequence of this is that at the steady state the final level contains a complete copy of everything. Poor write endurance makes flash SSDs poor participators in this sort of hierarchy.
/. -- the Free Republic of technology.
For many applications, what matters is price per IO operations per second. Sometimes price per IO operations per second per U of rack space or per watt is what matters.
Flash will beat hard drives there much sooner than it will beat hard drives in simple $/GB.
Flash has fast reads and slow writes. Sun is promising flash drives with 25% of the space reserved for bad block remapping, and a huge amount of supercapacitor-backed write cache. They are promising to release this in SFF HD form factor with a SAS interface, and SO-DIMM form factor with a presumably proprietary interface.
Even if Sun breaks its promise, as is typical for them, someone else will come out with the product.
is very dependent on the application. In particular it depends on the mix of linear vs random operations and the mix of reads and writes.
For 100% read applications SSDs tend to be similar in performance to hard disks when reading linearly, and a lot faster than hard disks when reading randomly. This shows up in linear read speeds of 100 MB/sec for a typical Flash SSD which is "close" to a hard disk. For random 4K reads, Flash SSDs can stomp any hard disk. Most disks are in the 10,000 4K read IOPS range where 15K SAS drives are in the 250 range or 40x slower. So for applications that are 100% read SSDs can be as much as 40x faster, although the average is usually in the range of 15x to 20x.
When you start writing to Flash things get interesting. Flash is really designed for large, linear, aligned, writes. With most drives, you can get maximum write throughput only if you write exactly aligned with the drives internal erase blocks. Thus you can write exactly 2 megabytes on exact 2 megabyte drive boundaries and get 100% of the theoretical write throughput of the drive. Unfortunately, no application acts like this, so you are at the mercy of the file system and Flash controller to turn your smaller, probably random, and probably mis-aligned writes into what the drive can handle. The net impact of this is that good Flash SSDs have 4K random write IOPS in the 120s which is 1/2 the speed of a 15K SAS drive. I have measured Flash SSD with 4K write IOPS with values like 135, 120, 64, 43, 24, 13, 4.0, and 3.3.
This is why Flash SSD performance is so hard to judge. The random write performance can suck up the available "drive time" and dig a system deep into dirty buffer flushing. We talked with one Dell laptop user that described their system becoming "unusable" while an Outlook indexing operations was randomly updating a big file. Unusable in this case was 2+ minutes for to bring up task manager.
These random writes also have a real impact on the wear of the drive. Every time you seek a write, you basically chew up a write/erase cycle, even if the write is only 4K long. If you look at a drive that claims 50 GB/day for 10 years, this is 50 GB of linear writes on exact erase block boundaries. If you write 4K randomly, the 50 GB really means 25,000 4K writes or 100 Megabytes of random writes.
The solution to this is to not write randomly to the drive. There are file systems designed for Flash that address these issues. These are typically called "Log File Systems". Unfortunately, there is no generally available file system really designed for performance. In Linux the LogFS options are really tuned for small memory small storage systems and for hardware where the flash chips are directly accessible. They do help drive wear a lot, but they are just not tuned for Gigabytes of space or database crunching performance.
Another solution is my companies product called MFT (Managed Flash Technology) which is a software block mapping layer that runs on the host. It gives you the random write performance benefits and wear benefits of a LogFS while allowing you to use whatever file system you wish. MFT was developed on 2.6 Linux and has been ported to Windows. With MFT, the same drives that do 25 4K random write IOPS usually measure over 10,000. The linear speed of the drive is still equal to a hard disk, but the random speed is now closer to symmetric with reads and writes. Thus jobs like updating databases can literally run 20x faster than the fastest hard disks.
In the end, Flash SSDs will find specific markets initially. I can say with certainty that they won't get used for off-line backups or storing/edit large quantities of HD video. But give them databases or file systems with lots of small files, and they can really smoke a hard drive.
The next step is to stop treating them as "disks". We tolerate the library and OS overhead of getting to a block on a disk drive because access times on disks are so long. But solid state memory devices can be accessed in microseconds. We need a different model for these devices.
With hot file adaptive clustering ( http://developer.apple.com/technotes/tn/tn1150.html#HotFile )
Basically, read only often accessed files are moved to the zone on hard drive where the access to files is fastest.
It would not be hard to adapt this behaviour to move the files onto SSD portion of the disk at all.
As the island of our knowledge grows, so does the shore of our ignorance.