Super-Fast Hard Drives

codders writes: "An Australian startup company, Platypus Technology, has launched a range of RAM-based solid state drives. These QikDRIVEs can offer sustained data throughput rates in excess of 110MBps and can be up to 8GB in size."

Comments

Re:Good for video capture?

[DrEldarion]

In response to "C)", the article states:

The QikDRIVE8 is compatible with an increasingly wide range of operating systems (OS) ranging from Windows, NT, Linux, Free BSD and Mac, and utilizes these OS's standard SCSI driver installation and management features.

Not sure if that's actually true or just marketing hype, but that's what they claim.

-- Dr. Eldarion --
It's not what it is, it's something else.

What's the big deal?

[jmroberts70]

People have been making virtual drives for years now. The only problem is that mobo makers don't create systems with that many DIMM slots (for obvious profiting reasons). What I don't understand is why we even bother with virtual memory at all. If you can afford to run a huge amount of RAM on your system, and never even come close to using all of it, why not disable using VM at all? Is there something I'm missing here?

Seriously, the present hard drive media has to go. It's the bottleneck of the entire system -and one of the few remaining moving parts in a computer. I'm looking foreward to the day when it's all solid-state.

Re:What's the big deal?

[Raven667]

Linux does this, and I am very thankful. (And remember kiddies, leaving a running Ethereal on a busy network while you go to lunch is a baaadd idea! The machine was unresponsive (although still responded to pings in 0.5ms) and constantly swapping for about a half hour after I got back, until the kernel decided to kill the right process.)

wrong solution

[jetson123]

It is generally better to add RAM as RAM on the main bus of the computer. That way, you have a lot more flexibility for how to use the extra memory: you can use it for caching, for applications, or as a RAM disk.

Note that RAM contents already survive reboots; it's the operating system that erases it (some systems take advantage of this fact for fast reboots). If you need power failure protection, you can also back up RAM that sits on the bus with batteries.

So, I think this card is a kludge, something that gives people a quick fix solution to a performance problem. For a quick fix, however, I would prefer a self-contained external box with a SCSI interface.

Re:Didn't We do This?

[orpheus]

The Platypus comes in 3 flavors: The QikCACHE without an external power supply The QikDRIVE with an external power supply, and the QikDATA that has 'the added protection of automated back-up of data to an independent hard disk drive in the event of complete power loss.'

And how pray tell, does the *independent hard drive* function in the event of a complete power loss?

The QikCache is 'card only' - when you turn the computer off, it gets wiped. But that's irrelevant, since the thread is specifically about the QikDrive.

The QikDrive is 'card plus external power supply' that allows you to turn your computer off (e.g. to make hardware changes) without wiping your data. It still isn't non volatile (it wipes if the power is cut) and it costs $9840.00 for 8GB. Quantum's pricelist is down right now, but I read it was $500-600 per GB or less than half the price of the QikDrive

The QikData has "mirroring capacity" and the "capability" to transfer to a Platypus HDD with a built-in UPS. This increases the price even more -- the HDD/UPS unit is extra. You also have to load the HDD back to the QikData before you can restore service. Not good.

A nonvolatile Quantum, on the other hand is not dependent on the power. Yank the cord, pull the Quantum, take it cross country, plug it in, just like a HDD. And you have instant high-speed access

So, while I have no strong feeling about either device, I'm hard pressed to see how you give the advantage to Platypus at all, much less $8K worth! No, I'm not being sarcastic. I really am curious why you gave the QikData the advantage. At all.

The Quantum is faster in "transactions/sec" (who cares about 'peak bandwidth'?), cheaper. And can be nonvolatile, straight out of the box without extra wires and components. (You can come crying to us when a technician seeing that the server is powered down feels free to unplug the QikData power cable. A UPS doesn't help, if you're not connected to it!)
_____________

Re:Ooooh

[mOdQuArK!]

Anyways, this would be great to stick your swap file on...

I guess if the motherboard (and operating system) supported it, then you could just GET that much memory (and get rid of the swap file entirely!).

Do they use cheap (and relatively slow) memory for this thing, or what? How much would 8GB of 133Mhz SDRAM cost if you bought retail?

Re:Some more info

[dragonfly_blue]

The original article detailing the press release (March 20th) is here.

The default configurations at those prices are $1538 for 512 MB (upgradeable to 1 GB) and $9840 for 4 GB (upgradeable to 8).

I'm curious; does anyone think that having an external power supply on the RAM drives make them worth the price premium over a software RAM drive?

Hopefully the prices will drop as the next generation of SDRAM factories comes online... ;-)

Why?

[HeghmoH]

I don't quite understand why you'd use one of these things. It seems to be that you'd get much better speed by simply putting all 8GB on your motherboard. Surely there are motherboards that can take that much memory, in addition to a 64-bit processor to address it all.

I guess the part about an independent power supply is useful. If the power goes out, a UPS is going to be able to power a dinky little card a lot longer than an entire server. However, if your server is under enough load that you need one of these things anyway, you probably have multiple safeguards in place should the power die. You could always keep a hard drive on standby and write your ramdisk to it when the UPS notifies the computer that the power's dead.

So, am I missing something? Is it less practical to cram that much memory onto a motherboard than I thought?

Re:Why?

[marcsiry]

Fast access to offline storage- video clips, etc.

Programs like Premiere and Avid Media Composer don't keep the entire video clips in memory- it would never work with 500MB clips being the routine. Instead, they are stored on RAID arrays or fibre channel-connected Storage Area Networks. Although fast, there is still appreciable time involved when it comes time to render the clips, as the information has to be pulled off disk, processed, and rewritten to disk. With a super-fast (i.e. RAM speed :-) disk, video editors and the like will see a large increase in speed.

Re:Why?

[Skapare]

Point me (URL) to a place where I can buy a motherboard with 8GB (or more) RAM support (and NOT with that RAMBUS crap) and a 64-bit CPU that is supported by Linux or FreeBSD.

Re:Application Structures, Volatility, Scale

[JordanH]

• • Application structure in general - Lots of applications know they're dealing with *files*, not memory, so you've got to give them something that looks like a disk, either by hanging it off a disk controller (using SCSI or IDE-like), or on a PCI bus (imitating a disk controller), or using RAMDISK drivers, or using hybrid memory/disk file systems like TMPFS.

  Again, this argument just dosen't hold salt.

Yes, but is it worth it's water?

While I'll agree that in general, these devices have limited usefullness, there ARE cases when an application developer can better guess at important usage patterns than can a caching algorithm.

For example, a given application might be optimizing indexes when there is no demand on it. When the application would be otherwise idle). In such a case, the caches would always be full of arbitrary indexes that are being optimized.

However, when a user actually attempts to use the application, 99 out of 100 times a particular set of pages might be called on first. In this case, it would be nice if you could guarantee that these pages could be accessed with minimal latency.

Having said this, I would, in general, rather have the directly addressed RAM (standard memory) rather than the SCSI bus RAM in my system. Even if there is an advantage to RAM drive optimizations, I'd prefer to have the flexibility of RAM drive software to turn segments of the directly addressed RAM into easily reconfigured RAM drives rather than inflexible SCSI bus RAM drives.

I also think argument that dropping a RAM drive in to give a boost to existing applications that this poster makes best, I think, is a good one. But, again, a RAM drive defined in software would be better here.

-Jordan Henderson

Required Microsoft Bashing Post

[Stoutlimb]

Microsoft should buy up this company, and then make these drives as cheap as possible for everyone to use. Reboot time would would go down significantly! They would save billions in not having to make their code more efficient.

"The hardest thing to understand is the income tax." - Albert Einstein

What goes around ...

[Virtex]

I remember seeing an ad for a laptop computer a few years back with a solid state hard drive. The idea was that data could be accessed faster, and it would not drain the battery as fast. The only problem was that the drive was only 4 MB in size. It seemed like a good idea, but at only 4 MB, it also seemed like an idea that just wouldn't work.

Now we have these QikDRIVES capable of holding up to 8GB. Finally, enough space to be useful. But because they are using ECC SDRAM, the prices are going to be too expensive for all but the most serious of consumers. It's generally been accepted that hard drives are slower than RAM, so why not use cheaper RAM that runs at, say, 60ns (like the old EDO RAM)? We would still probably want to use ECC or something similar for data integrity. Would this cut the price sufficiently to make it attractive for the average person? Obviously, it would still cost more than a conventional hard drive, but hopefully not nearly as much as the current line of QikDRIVES.

Now I'm not saying to dump the existing line, since they will still be attractive to those who need the high performance. I'm just thinking of how to reduce prices. And over time, the prices on these things can be expected to drop anyway, perhaps even making them commonplace. Does this sound plausible?

--

Hardware == Hard drive?

[yerricde]

And a floppy disk drive is also hardware, but that doesn't make it a hard drive.

Re:Instant on-

[yerricde]

For me the main thing is having enough *reliable* space to store my digi-junk.

For you the solution would probably be a four letter word: cdrw.

Before anyone gets carried away...

[orpheus]

Before anyone gets carried away, please remember that these are volatile RAM drives.

The Platypus overcomes the HDD's primary liability (read/write latency) at a serious cost to the HDDs primary function: reliable storage. Note that it doesn't even have an on-board battery. It simply has a separate external Power supply and (optional) UPS

While a UPS is wonderful for keeping my system running, it's much less reliable than it needs to be if an outage (or office idiot kicking the plug out) means I lose *all* my data (sales for the day, etc.) In a sense, the platypus drive is not much stabler than having 8GB of system RAM and *no* HDD ["not much" is relative. The MTBF of a UPS is orders of magnitude less than a good HDD)

I doubt the usual high reliability filesystems could maintain a RAID/HA type redundant backup to disk precisely because the RAM HD is so much faster than the disk. It would be like having a scribe backing up your HD to quill-and-scroll -- the more you utilize the tremendous speed of the RAM HD, the farther behind the disk will fall (and thrash).

It's a nice product (though hardly a new idea), but I see it having limited application (e.g. as a HD accelerator in some server applications)

Perhaps someone can do a hardware workaround using an intermediate NVRAM between the SDRAM HD and the hard disk, using principles borrowed from both cache technology and High reliability file systems. But it'll take a bit of work.

Is there already a solution out there? Or is this essentially just a giant unidirectional HDD cache, good for serving up data faster than an HDD, but not good for critical rewritten data?
_____________

Re:Good for video capture?

[Raven667]

According to their website it "works like a SCSI drive" and has Linux support. My guess is that it has its own SCSI controller on the card, which may have been tweaked to support the high bandwidth (110MB/sec) of the RAM. What does Ultra2 SCSI support, 80MB/sec? This might be useful as an external SCSI device but you would lose performance, I think.

Re:Eliminating Latency is the big win

[jafac]

The thing I don't understand is, why-oh-why do they insist on using like, the highest-cost RAM available for this application, when there's plenty of 4-meg simms out there from old machines and memory upgrades that are not being used. Why can't someone create a drive like this that just has arrays of empty simm slots, so we can get this currently worthless commodity, and put it to good use?

I just remembered this old Metallica song. . .

I've used a prototype (not from Platapus)

[darial]

Unfortunatly, I'm under NDA, so I can't tell you everything about it, but I have a fairly good grasp on the performance of this class of device.

The biggest problem with 32 bit machines is not the 32 bit int, which is really sufficient for most things, it's the inability to address more than 4 gig of memory. This provides a relatively clean solution to that problem by using this device as swap. The burning question, of course, is performance... how much worse is it than on board memory?

The answer is, of corse, it depends: If you are in a single process enviroment, the time it takes to swap pages is somewhat killer, because the machine justs sits on it's ass while the DMA moves the block. Now, don't get me wrong, it's a lot better than disk, but it's not like real meory.

However, on multi-process machines like servers, it's great. There is a delay for the page swap, but the other processes keep the cpu busy and the DMA keeps the bus busy. Since throughput is more important than response time, this is almost as good as onboard ram. But, you say, this is MORE expensive than real RAM? not really... for an app like this, it will be an smp machine anyways, and the difference in cost between comodity x86 parts and a 64bit+8gig-uberboard setup from a proprietary vendor is so great that you could buy one of these things with the spare change. This could easily save many tens of thousands on certain types of server projects.

Re:I've used a prototype (not from Platapus)

[soldack]

"The biggest problem with 32 bit machines is not the 32 bit int, which is really sufficient for most things, it's the inability to address more than 4 gig of memory."

Actually Intel based systems can addres up to 64 GB of memory using Physical Address Extensions (PAE) for 36-bit addressing.

Hardware needs to be 64-bit compliant and/or support Dual Address Cycle (DAC) to address memory above 4 GB.

"because the machine justs sits on it's ass while the DMA moves the block."

If you are using a busmaster PCI device (which any SCSI or Fibre Channel card is) the card itself does the DMA. Most modern operating systems support non-blocking I/O. This means that a read/write call can get a pending response. The caller can than go about other things or wait on an event to signal the I/O completion. If the caller waits, the OS can run another thread or process until the event is signaled. If the caller doesn't wait, it can still be interrupted when its quantum expires.

I have worked on a Fibre Channel/RAID adapter that blows this thing out of the water. 110 MBps? In vanilla direct connect mode (no raid), our card did 190 MBps and 20,000 I/Os per second. With RAID and a 64 MB cache, we can beat that. 8GB? We worked with 1TB to 10TB databases. By putting only up to 8GB on one PCI card you are really limited. 64-bit, 66 Mhz PCI supports up to 528 MBps. Five of their cards on one bus would give you only 40 GB at a transfer rate of 105.6 MBps per card. You can add busses but that isn't cost effective. On other hand, you can put nearly unlimted storage on one of cards. For the best speed you can spread it out over multiple ports (our card has two) and multiple cards. 3 cards on a 64-bit, 66 Mhz PCI bus can still hit 176 MBps each (2 cards can hit 190 MBps each) with MUCH more storage.
This thing is too small and too slow for their target market of enterprise computing.

I've been thinking about this for years!

[piku]

Though my idea is a bit different. All it involves is a few gig of normal ram and loading the contents of the harddrive into it on boot. All changes would be done in ram, and then on shutdown it would copy the entire thing back onto the harddrive. Super fast access, 30 minute startup/shutdown. Good tradeoff if you have a long uptime.

Re:I've been thinking about this for years!

[vectro]

This is basically just a system with lots and lots of cache. So assuming you are on a system with more memory than disk space, any modern operating system will do this for you, commiting changes to disk occasionally.

Of course, this would have to be a 64-bit architecture to be able to address that much memory, and you would need an OS that supports that much memory.

Re:If it costs $600, I'll have one.

[Xenu]

As a cheaper way of speeding up hard disks, why don't drives have multiple heads per platter?

The problem is keeping the head aligned with the track on the disk. This is difficult enough with a single active head. As soon as you introduce multiple active heads, you have the problem of keeping multiple heads independently aligned with their tracks. Putting two heads on a single positioner arm isn't going to work because of thermal expansion/contraction of the positioner arm and mechanical errors with alignment of the positioner arm and the disk platter. You need independent positioner arms, each with its own servo system and read/write electronics. This is very expensive. Multiple heads on a positioner arm and head per track disks used to be common in high-performance disk drives when track densities were much lower than today. I used to use a computer that had a 5 MB head per track disk drive, it used up a whole 19" rack.

Re:It's about time

[Detritus]

One method to improve the performance, even with current technology, might be to get rid of the well known structure of hard discs though. As it has been for years now discs are platters with a fixed logical structure. 512 sectors, N sectors per cylinder, N cylinders per platter.

It has already happened. The heads, tracks and cylinders that you see in the BIOS setup screen on a PC have no basis in reality. That is just a software compatibility kludge for PC operating systems. If you look at the SCSI specification, you will find no mention of heads, tracks and cylinders, the disk is addressed as an array of logical blocks. IDE drives can use logical block addressing, similar to SCSI, or heads, tracks and cylinders. Due to the use of techniques such as Zoned Bit Recording (ZBR), the number of sectors per track is not a constant. The IDE drive translates the physical sector address provided by the CPU to/from an internal sector address that reflects the actual layout of the drive. I've seen SCSI drives with larger sector sizes but they are rare.

Why does it require THEIR sdram memory?

[Skapare]

Why can't it use generic sdram?

Re:Good for video capture?

[Heartsbane]

Not quite video cap/editing but a UK mag called Computer Arts reviewed a solid state drive a few years ago. I think it was It was external and had a capacity of around one gig.
They tried it as scratch-disk for Photoshop with, IIRC, a 2 or 3 hundred meg file and said it was like working in RAM.

*sigh* Yet another thing for my wish-list.

---
"When I was a kid computers were giant walk-in wardrobes served by a priesthood with punch cards."

Application Structures, Volatility, Scale

[billstewart]

The main reasons are

• Application structure in general - Lots of applications know they're dealing with *files*, not memory, so you've got to give them something that looks like a disk, either by hanging it off a disk controller (using SCSI or IDE-like), or on a PCI bus (imitating a disk controller), or using RAMDISK drivers, or using hybrid memory/disk file systems like TMPFS.
  
• Database-specific needs, particularly committment - DBMSs really need to know that when they've written something to stable storage, it'll stay there unless they change it - crashing the machine or losing power will trash things in regular RAM, and that just doesn't cut it. Sticking a bunch of RAM in a SCSI shoebox with its own stable power supply and maybe an automagic copy-to-builtin-disk gives you the security you need, and 8GB really *did* hold a large database not very long ago :-) Putting it on an imitation disk controller board required being more careful about armoring it, but you might be able to do it. And building a stable storage device that doesn't need to wait 5-10ms to commit data is a big big performance win.
  

There are also scaling issues - motherboards always have some limit on capacity, whether it's address lines or card slots or whatever. With Ultra-Mega-FooBar-SCSI, you can hang 8-16 of these things on the bus if you need to. Living on a bus lets you design boards for your specific application, and isn't limited by the design tradeoffs and compatibility requirements of a general-purpose computer, just by the size, power, and cooling of a shoebox or 1-2U and the creativity of the designer. Can *your* desktop machine address or even hold 8GB of RAM? Mine can't.

Re:Application Structures, Volatility, Scale

[vectro]

Application structure in general - Lots of applications know they're dealing with *files*, not memory, so you've got to give them something that looks like a disk, either by hanging it off a disk controller (using SCSI or IDE-like), or on a PCI bus (imitating a disk controller), or using RAMDISK drivers, or using hybrid memory/disk file systems like TMPFS.

Again, this argument just dosen't hold salt. If you want to have something that fast, then just provide lots of RAM and your OS will use it as a cache. There's literally NO reason to use a ramdisk except as an initial boot loader if you're using a modern operating system.

Some more info

[Chairboy]

I read about these in March. Here's some info from then that might have changed in the interim:

$1,538 for a QikDrive1 with up to 1GB storage.
$9,840 for a QikDrive8 with up to 8GB storage.

The QikDrive is on a PCI card, but has its own internal power supply for data security. Presumably to keep the drives from being wiped by a system power failure.

They support between 15,000 and 20,000 I/O transactions per second (versus 200-300 for Winchester-style drives)

Re:Some more info

[ZikZak]

Oh, good! I was worried they would be expensive or something :^P

Does anyone know

[tcd004]

if these would outlast a regualr HD. It seems to me ramchips might degrade even faster than HD wafers.

Just think how fast Metallica MP3's can be downloaded to these babies! tcd004

Here's my Microsoft parody, where's yours?

Actually, it's quite cheap.

[billstewart]

Sure, it costs a lot more per GB than rotating machinery, and I wouldn't buy one for home, but in a business context it's a really cheap deal if it works well. You can get a substantial speedup for less than one day's consulting fee for a database wizard. (Of course, if it's not a good match, you might end up spending a few days's consultant time to make it work :-) Also, the performance impact may save you buying another computer, which would cost at least as much for a production server.

Re:Nice data throughput

[bonehead]

I don't see too big of a use outside of the overpriced mega-server market, though,

This thing would seriously rock for video capture. I'm currently having to capture video in compressed form to avoid hard drive bandwidth issues.

Old stuff

[InsaneGeek]

Solid state disks have been around for years, they aren't offering anything a bit revolutionary or even a best of breed product. They even take out the functionality of a SSD by making you plug it into a pci slot, instead of a standard drive connection. Maybe if they've SIGNIFICANTLY reduced the price it might not be that bad, but that's a big if.

Check out solidstate.com, mti.com, etc. they have much better solutions than putting a card into your computer.

I checked out SSD's last year to see about SAN integration, but the cost if VERY prohibitive. i.e. 90k for a 4 gig disk with a fibre channel connections (of course that was battery backed up, disk backed up, etc). If you are running a big data base/warehouse they can become very useful, they appear to the system as a regular drive, no drivers, etc. I know of a couple of companies who do a raid set over multiples of these (think 10x4 gig striped SSD's to do big database billing then think price).

Legato NFS Accelerator Boards for Suns

[billstewart]

Forgot to mention - Legato used to make an accelerator board for Suns that was basically a meg or two of battery-backed RAM on an S-Bus card, with some appropriate driver support. As with databases, being able to commit a write without waiting 5-10ms for mechanical latency was a big performance win for NFS, and a meg or two was enough buffering to do the job. It could also be used for database journaling files, and similarly helped a lot. Keeping a couple MB of SRAM or DRAM alive doesn't take hulking UPS batteries - a little NiCAD or lithium cell can keep you alive through pretty long power failures, and the drivers were written to write any leftovers to disk at boot time if they hadn't been written already.

Why so slow?

[cperciva]

110MBps? Come on, even PC100 RAM can transmit 800MBps, and DDR and RDRAM get up into the 2-4GBps range. Sure, flash memory is slower, but if they are building gigabytes of it they should be able to use interleaving to speed it up.

I think that at very least they should be able to run at the 533MBps maximum speed provided by 64-bit 66MHz PCI.

Re:Why so slow?

[soldack]

Actually, it gets worse. Because they only offer 8GB per card, you have to use multiple cards to get to any kind of enterprise size. 5 cards gives you only 40 GB at 105.6 MBps. Yuck. A good Fibre Channel card can give you 150-200 MBps over 1TB or more. Adding more PCI buses gets expensive and relies on OEM guys to make a special box for you. Check out the cost for multiple 64-bit 66 Mhz PCI Bus systems. This is not a good solution.
Good cache managment on a RAID card can get this kind of performance per RAID drive using much less RAM. You computer doesn't have 64 MB of L1, 2, or 3 cache, does it?

Re:Wow! Something totally new!

[jaystile]

Solid State Drives have been around for awhile. The only problem is the extreme cost. The fact they are releasing a range of SSDs is just another step in computer evolution. I remember reading an artical (I can't remember if it was from /. or not) where they are using these metallic doughnut rings that are about .5 microns in diameter that hold a magnetic charge when you shoot a pulse down through the hole. And it stays when you power off your computer, so when you turn it back on, it's instant on. No more booting. If anyone remembers where this site is please give me a post. I also think I remember seeing that it held about 100GB/in^2. So we have a lot more to look forward to than just SSDs.

Re:Ooooh

[vectro]

Actually, you still want a small amount of swap. Most machines have processes (eg at and cron) that sit around for a long time doing nothing. There is negligable performance loss to swapping these apps out to disk, leaving more memory for important things. So you almost always want at least a small amount (100MB) of swap.

Re:Using RAMdisks for security, scratch space

[vectro]

I'm well aware of the use of a RAMdisk as an initial bootloader. I've actually got a neat set up here where I modified the redhat install disks to boot a system over NFS after configuring and loading the networking and NFS modules. So I know all about ramdisks as a boot loader.

But as for the scratch space, I believe there is a function to request a block of memory never go to swap. I know linux has this feature, I thought windows did too.

Good for video capture?

[Team_Z]

Sounds like these things would be great for my work, which is video editing. The problem with many drives I've used is that they get trashed by capturing video constantly, slamming data on to those drives for sustained periods of time can trash them in a few short months due to the constant accessing. Any of you who've used a Media100 or other hardward packages similar to it may know what I'm talking about.

My main concerns are:
A) Is this here to stay, or is it going the way of the SparQ?

B) Is it going to be cost-effective? Granted an Art Director like myself can just get the company to pay for the drive as part of a new system or upgrade, but many freelancers would have trouble with a super expensive drive. . .

C) Are they really going to be worth using, or am I going to have constant trouble getting the damned things to work on my Operating System(s) of Choice[tm]?

Eliminating Latency is the big win

[billstewart]

Devices like this have been around for a decade or more, but as memory keeps getting radically cheaper they become more interesting. The big performance win isn't usually how many MB/s throughput you get, though that's nice, but it's the elimination of rotational and seek latency which can be 5-10ms - a long time for a database transaction, and even with journaling filesystems or datastructures you've got to deal with some of it.

Be sure to use a good UPS with the things, and make sure your powerfail shutdown procedures work well.