Slashdot Mirror
TB-Sized Solid State Drives Announced
prostoalex writes "Several companies have announced solid state hard drives in excess of one terrabyte in size. ComputerWorld describes one from BitMicro that's just 3.5". Their flash drive will support up to 4 Gbps data transfer rate. From the article: 'SSDs access data in microseconds, instead of the millliseconds that traditional hard drives use to retrieve data. The BitMicro E-Disk Altima 4Gb FC delivers more than 55,000 I/O operations per second (IOPS) and has a sustained data transfer rate over 230MB/sec. By comparison, a fast hard drive for example will run at around 300 IOPS.'" Ah, the speed of tech. Seems like only last month we were talking about 500GB drives.
The Texas Memory Systems datasheet claims 24 GB/second of random sustainable data bandwidth which is much higher than the Fusion IO card but it looks like they are serializing this possibly across multiple drives. They also claim higher (3.2 million) operations per second.
The BitMicro drive is groin grabbingly amazing in size but claims only 55k operations per second & sustained data transfer rate over 230MB/sec.
So what I would wager is that PCIe might provide more throughput than SATA but don't quote me on that. I'm interested to see where this goes & also curious to see whether we continue dumping drives on channels like the Texas Memory solution or if it just goes back to a server with a ton of PCIe slots on it and hot pluggable card swapping for 'drives.'
Worth revisiting is the fact that Fusion IO claims to be releasing the cards for sale next month. As we all know, sometimes it's just a case of who gets to market first that wins in the technology world.
My work here is dung.
The TMS link is for a 9U rack of non-volatile DDRRAM, consuming 2.5KW and weighing up to 720lbs, so not quite suitable for the desktop.
The BitMicro article goes on to say that the maximum capacity in a standard 3.5"x1" format is 640GB, so requiring around 2.5" for the full 1TB.
This is Slashdot, so we don't expect facts in the summary to be correct. However, this is still amazing progress.
Any sufficiently advanced bug is indistinguishable from a feature.
The linked to press release for TMS systems are not a single drive. They are a half rack sized array. Dont try and put one in your desktop anytime soon.
:-)
Their systems have been in use for years by folks who need speed at any cost.
Now, the BitMicro drives... those look interesting. I wonder if I can slot them into my StorageTek 6140
The big feature here is the included Infiniband support. Without digging real far in to the specs if this array supports RDMA it would make a very nice shared memory array for a grid type implementation.
This comes up every single time flash drives get mentioned on Slashdot. Go search around and you'll dozens of posts in every article asking and answering this question. The short answer is that with the wear levelling used on all modern flash drives they work out at at least an order of magnitude more reliable than current HDDs despite the write limit.
Spelling mistakes, grammatical errors, and stupid comments are intentional.
So, it's a giant ram disk with either flash or hard drive backup. http://www.superssd.com/faq.htm
When our name is on the back of your car, we're behind you all the way!
Apparently this works on Windows XP too: http://www.windowsxlive.net/?p=1337
RamSan-400
The starting capacity of a RamSan-400 (32GB) is $35,000. It includes:
-32GB DDRRAM storage
-one dual-ported 4Gb Fibre Channel controller
-hot swappable RAID 3 hard disk drives
-hot swappable and redundant power supplies
-redundant battery and fans
-IBM Chipkill in memory (redundant RAM)
-1 year return to factory warranty
Each additional 4Gb FC controller is $3,000 (up to 4 in each chassis).
The RamSan-400 can upgrade in 32GB increments for $18,000 (up to 128GB).
RamSan-400 (64GB) - $50,400
RamSan-400 (96GB) - $65,800
RamSan-400 (128GB) - $81,200
RamSan-500
The 1TB base-level system of a RamSan-500 (1TB SLC NAND Flash, 16GB DDR) is $200,000. It includes:
-one dual-ported 4Gb Fibre Channel controller
-hot swappable and redundant power supplies
-redundant battery and fans
-1 year return to factory warranty
The 2TB base-level system of a RamSan-500 (2TB SLC NAND Flash, 32GB DDR) is $300,000. It includes:
-two dual-ported 4Gb Fibre Channel controllers
-hot swappable and redundant power supplies
-redundant battery and fans
-1 year return to factory warranty
The RamSan-500 can upgrade DDR Cache.
-16GB to 32GB is $10,000
-32GB to 64GB is $20,000
Each additional 4Gb FC controller is $3,000 (up to 4 in each chassis).
SmartBox
http://photofile.ru/photo/fishki_net/3273658/70316241.jpg
Rich
It'll take only about a year or so; Samsung 8Gbit Flash chips are $9 on the spot market at the moment.
Though the interface for connecting eighty of them onto a single SATA channel wouldn't be completely straightforward just from the point of view of I/O pins; each chip presents eight data pins and nine control pins, so you need a total of about 1500 pins on the disc-controller ICs.
For the early market you'd probably use FPGAs, you'd need six XC3S400A in one of the larger packages to get enough I/Os, which would be another couple of hundred dollars; the master one can speak SATA directly and do a bit of buffering, and using this baroque configuration means you get the full data rate from each chip so reads are very fast. Say $1000 for 80G SATA-attached - but that's making mass-manufacturing assumptions, which are clearly false if what you're manufacturing are 80G disc drives for $1000.
Five hundred dollars for 80G SATA-attached sounds a reasonable sort of pricing once Samsung have got one more generation of Flash factory online, and once the devices are more plausibly consumer items.
Actually,
1. Even for low level disk access, that hasn't been so since the days of MFM hard drives. Nowadays everything uses LBA (Logical Block Addressing). Meaning that when the computer wants a certain sector it tells the hard drive, quite literally, something like "give me block #13526".
2. As a side effect, this already allows the hard drive to remap around bad sectors. If you read blocks #13525, #13526 and #13527, you might get the middle one from a whole other position than the other two because it was remapped.
Disk defragmenting is based on the assumption that contiguous logically _probably_ means contiguous physically too, but there is no guarantee that it's actually so. _Probably_ the HDD won't remap when there's no need, but again, it wouldn't tell you anyway.
3. For _filesystems_ doubly so. Even the FAT in DOS 1.0 didn't work with tracks and sectors, it worked with block numbers. The translation to cylinder, head and sector was made at a whole other level to actually read or write the data. But the filesystem didn't contain any reference to those.
E.g.: a 1.44 MB hard drive image still works flawlessly when copied to the first 1.44 MB of a CD. (That's how bootable CDs work. They have a floppy image at the start, and it's really booting that.) The FAT contained no references to cylinders and heads, so the exact same image works just as well off a CD.
4. Well, it's not that new a problem. You know those USB memory sticks one can buy? Or connecting a Flash-based MP3 player to your PC and copying files on it? Those tend to be formatted as FAT. So there you go. They don't have to invent anything new for an internal SSD. They already did it on other devices.
A polar bear is a cartesian bear after a coordinate transform.