Slashdot Mirror

That doesn't seem possible. Even the current generation of FusionIO's high-end products are specified at 2.7 GB/s read and no more than 330K IOPS (4K random read). (As an aside, if you price hunt, you can get the PX600-2600, which pushes that many read IOPS, for about $23,500, or just over $9/GB).

Put another way, that latest generation of products use PCIe 2.0 x8, which is 40 GT/s. After the 8b/10b encoding, that's 32 GT/s; so before any PCIe protocol overhead, much less anything application-specific, they couldn't possibly be going any faster than 4 GiB/s.

A fast hard drive can do several hundred or with short stroking on a high RPM drive maybe even a thousand random IOPS (input/output operations per second).

A fast SSD can do 90,000+ random IOPS - or several hundred thousand for direct native PCIe drives that don't have to go through a SATA layer.

For what SSD's do well which is randomly accessing data, they aren't 5 times faster than HD's, they are 100 to 1,000 times faster at RANDOM ACCESS for a SATA connected SSD.

There is a single board Fusion IO card that can do 9 MILLION IOPS which is roughly 10,000 times faster than the fastest hard drives.

Not in real world use. There are no 1M IOPS SLC SSDs (single drive), but there are plenty of 100K IOPS MLC SSDs.

As a matter of fact, this seems to show that with the exception of the Fusion-io ioDrive2 SLC variant, all the top-performing single drive SSDs are MLC. And, the MLC variants of the ioDrive2 are only about 10% behind the SLC variant.

You can see from the Wikipedia article that what truly affects final throughput is the bus width and number of channels of SSD controller, just like I said. The fastest systems are just many MLC SSDs connected to a very fast bus.

If this were even close to true, large corporations would not use NVRAM technologies to back their incredibly critical data stores. That "spinning rust" in a mid-sized 8-drive RAID-10 array can deliver roughly 2000 operations per second. One 2.4TB FusionIO drive for example? over 500,000. There's not even any comparison here. The size and cost of the SAN you'd have to buy to come even close to those numbers using traditional platters is on the order of multiple racks, compared to a single PCIe card.

I thought your information might be a little out of date, since I remembered hearing about those 500k IOPS PCIe card-drives a while back, so I did a bit of googling for current stats. It seems that back in March FusionIO managed to get one a 365GB ioDrive2 to pull 9.6 million IOPS using a custom filesystem and some other software tweaks. However, looking at their products for sale section, they're currently only advertising 1.3 million for their much larger (and probably still insanely expensive) 10.2TB ioDrive Octal, so your comparison vs. spinning rust is only potentially an order of magnitude more impressive than previously stated.

SSDs are slow in that they rely on old school disk protocols like sata. Sure, you'll get better performance than spinning disk. But if you want screaming fast performance, you should look at flash devices connected through the PCIe bus.

Products from Fusion IO would be an example of this. Apple Mac Pro would be another: "Up to 2.5 times faster than the fastest SATA-based solid-state drive".

You can look at Fusion I/O PCI cards. Not cheap at all, but certainly fast.

Triple the performance is pretty good for cache. I see that too, particularly with the newer SAS RAID controllers with 1GB of flash-backed write cache. A decent modern SATA attached SSD can do something like 70x the IOPs and 20x the bandwidth of a spinning disk though, and also sees additional benefits from RAM caching. The PCIe attached versions can do well over 1000x the IOPs of 15K 6G SAS. No joke, one PCIe attached internal flash drive can outperform on I/O even the most optimally configured array with the 1000 of the fastest available spinning disks (which typically would take at least 3 full racks and 25 KW and have a drive failure every week). And the latency is down near 26 microseconds in the worst case, where a cache miss is best-case 2000 microseconds of latency on a spinning disk.

Solid state: it's just faster. It also costs a bit more, but not offensively so. RAM is getting a lot cheaper these days (about $8/GB for the registered 8GB DIMM) so if you've got a problem set that fits in configurable memory and doesn't require the reliability of static storage during the problem set, RAMDISK is definitely an option again for some. Frankly these RAM prices look crazy cheap to those of us who were praying for the day that RAM dropped below $100/megabyte back in the day. I've paid over $300,000/GB for RAM ($300/MB) - and it was much slower RAM too. Going over the math in my head, apparently I once also paid $2,500,000/TB for HDD storage ($300 for 0.00002 TB) - out of my own pocket as a consumer. That's just crazy.

The question comes down to what is your time worth? What can you do with storage this fast? Does your data fit on the SSDs? Are there other benefits? Everybody has different needs. The available technology is outpacing our needs by a good bit these days. You now can fit all of Peak Twitter into the storage, network and processing capacity of one off-the-rack server - though of course you would use three and geographically isolate them for HA if you were Twitter. These new technologies enable us to do things we couldn't do before - or alternately they allow us to use software that's so offensively bloated it robs us of all the benefit of technology's advance.

Subbing an SSD for your laptop or desktop's boot HDD will get boot times down to mere seconds even in Windows. Anyway, I'm rambling. Time to click "submit".

Yeah, about that 4TB limit, I think these folks will be surprised that their 5TB and 10TB drives won't be possible in the next few years....

No, I do believe that's 2TB of directly addressable RAM as one contiguous chunk in one 4U server like the HP DL580 G7 for example, and 1M IOPS consisting of four devices in one server each capable of 250K 512 byte IOPS, and striping in software to deliver the full 1M IOPS in one server. Specifically for storage the IODrive Duo SLC. You can actually do 2M IOPS in that box with 8 devices, but that seems overkill for most things. There's esoteric stuff out there that uplifts this stuff to 20x even that, but uses flash storage as a sort-of second tier of RAM and it's too experimental to consider for enterprise use - and the box becomes a storage only node rather than a general purpose device. You can do some more traditional SFF SSD storage as well if you like, for a second tier of storage. They're doing memory mapping stuff with Infiniband too now I understand, but I didn't figure that either.

This is commercial, off-the-shelf stuff now. Clustering would give you multiples of this naturally, but not linearly and with the concerns that you mention. The next generation of servers is due out any day and will use Load Reduction DIMMs and PCIe 3 to double up both the installable memory and the storage I/O potentials. Industry standard servers are getting pretty hardcore.

Why would a billionaire stand in line all night long to get a cellphone that his company sells?
you know he works for Fusion-IO, right? that he hasn't worked at apple for 20+ years?

What I mean with lackluster on x86 etc is that I/O is still sequential bus limited, and even with DMA etc, the CPU STILL has to do some of the I/O shuffling gruntwork.

This discussion has come up in numerous places over the last few years and is basically false. The majority of the modern x86 peripherals have as much if not more of intelligence than channel processors. For example fiber channel and SAS boards from qlogic/emulex/etc have full blown processors on them running firmware that handles all of the fiber channel protocol and a large part of the FCP portions. Leaving the CPU's to do little more than specify via SCSI CDB's and target ids which data blocks get moved where. Once the operation(s) are complete the board interrupts a CPU. These boards maintain all the connections, and keep track of tens of thousands of simultaneous IOs. The CPU usage to transfer 3GB/sec to/from disk in our setup is less than 1% and a large portion of that is our application sending messages to-from the OS. Its the same with inifiniband, as the protocol is handled by the adapter, leaving the CPU to do little more than trigger the remote operations.

Combined with the fact that PCIe now includes peer to peer as part of the standard means that you can actually do IO between devices with out even the memory subsystem getting involved. This is how GPU's are doing SLI.

Anyway, I think the original discussion was more about how intel was intending to displace the RISC vendors, aka the power systems not the mainframes. Either way, I think my original point stands, as i'm betting the system your talking about is well into the 7 figure range, or roughly two orders of magnitude more expensive for what is probably only one order of magnitude faster than a single node in our cluster. As our application has nearly linear scaling for node counts in the few dozen range we are an example of an application that probably gets similar (if not greater) IO and processing performance out of cheap Intel hardware.

BTW: Texas Memory Systems makes some cool stuff, and systems like http://www.fusionio.com/products/iodrive-octal/ do a lot to move cheap intel hardware into places that traditionally required big iron.

OCZ was hardly the first to do direct flash-PCIe storage(though some of their earlier products were just a disk controller and SSD on the same card, they have native ones now). They do have the advantage of being one of the vendors of PCIe-SSDs whose prices are remotely accessible, and who are available through enthusiast channels.

Most of the other players are basically in the business of making people's Big Serious Expensive databases run faster, and their prices and "if you are serious, please call our sales department" distribution style. You probably don't want to know what one of these or these cost, and you won't find them at newegg...

That's rather limiting. There are PCIe attached solutions that consistently read/write at more than 6GB/s rather than 6Gb/s - like for example the ioDrive Octal. It can have far more storage than your limit - ten times as much on one card. That thing has a serious 48Gbps serial read bandwidth, sustained, and you can configure many PCs with eight or sixteen of them. This is only one of many. There are actually some applications that strain against the limitation of this bandwidth.

The good rule for a thumb rule is that it should be general enough to scope its use. By bragging your rule on slashdot you've only defined the limits of your own vision and use. You're advertising that you're either "old school" or you're limited by some vendor's products, or late. Not a good place to be.

Here's a good rule of thumb: A storage drive can never be so fast, nor so capacious, nor responsive enough to serve every storage customer's need - but on a clear disk you can seek forever.

Do you see what I did there?

anything much above 10Gbps and you're going faster than most storage devices can currently handle anwyay,

Not true for long. Infiniband EDR 12x is 300Gbit/sec. It's only a matter of time before that speed hits the desktop. The fastest single internal device you can buy currently goes 6Gbit/sec. You'd need a cluster linked via Infiniband to reach 300Gbit, probably around 9 nodes with 6 cards per node. It's definitely attainable.

Uh, connecting a RAID array? Current x4 SAS provides 24Gbit per port. Thats ~2x what Light Peak/Thunderbolt can provide. Plus, I don't see any converters to FC so that I can attach to a SAN, or a nice place to put one of these http://www.fusionio.com/products/iodriveoctal. BTW: If your RAID can't saturate 6Gbit SATA then you might think about a different array.

there isn't a drive in consumerland (spinning or otherwiwse) that can use a full sata channel on its own.

What about the ioDrive? They have to use PCIe because SATA isnt fast enough.

(even sata150 is faster than ssd's are, sustained).

I think you're wrong. From http://en.wikipedia.org/wiki/Serial_ATA

As of April 2010 mechanical hard disk drives can transfer data at up to 157 MB/s, which is beyond the capabilities of the older PATA/133 specification and also exceeds a SATA 1.5 Gbit/s link. High-performance flash drives can transfer data at up to 308 MB/s which exceeds a SATA 3 Gbit/s link.

However the smart hard drive vendor would realize that spinning platters are headed out the door, and that they should invest in solid state technology, lest they be left in the dust. There's nothing really stopping the availability of high capacity SSDs except cost. You can already get 1.28 TB SSDs with insane speeds (1.1 GB/s read, 1.5 GB/s write), if only you're willing to pay the cash. As prices come down, there will be no reason to get a spinning platter drive. Notice how all the SSD makers are not the big HardDisk makers. They should be shaking in their boots, because a large part of their business is going to go away within 5 years. If spinning platter makers don't change something soon, their market is going to be reduced to a small fraction of what it was.

While the ioDrive may offer great performance, I hate their marketing.

http://www.fusionio.com/products/iodrive/

• Not an SSD - easily outperforms dozens of SSDs and a single server
• From 80GB - 320GB of enterprise-grade, solid-state Flash

"It's not SSD + RAID, it's solid state memory in parallel channels!"

No, it's not X25-M's on an Adaptec card. However, it is NAND flash with a bunch of parallel channels. It's the exact same idea behind SSD + RAID, it's just above the level that you'll get with "regular" SSD + RAID.

I'm kinda fed up waiting for the SSD manufacturers to get their act together. There's just no reason for drives to be only 10-50x faster than physical drives. It should be trivial to make them many thousands of times faster.

I suspect that most drives we're seeing are too full of compromises to unlock the real potential of flash storage. Manufacturers are sticking to 'safe' markets and form factors. For example, they all seem to target the 2.5" laptop drive market, so all the SSD controllers I've seen so far are all very low power (~1W), which seriously limits their performance. Also, very few drives use PCI-e natively as a bus, most consumer PCI-e SSDs are actually four SATA SSDs attached to a generic SATA RAID card, which is just... sad. It's also telling that it's a factor of two cheaper to just go and buy four SSDs and RAID them using an off-the-shelf RAID controller! (*)

Meanwhile, FusionIO makes PCI-e cards that can do 100-200K IOPS at speeds of about 1GB/sec! Sure, they're expensive, but 90% of that is because they're a very small volume product targeted at the 'enterprise' market, which automatically inflates the price by a '0' or two. Take a look at a photo of one of their cards. The controller chip has a heat sink, because it's designed for performance, not power efficiency!

This reminiscent of the early days of the 3D accelerator market. On one side, there was the high-performing 'enterprise' series of products from Silicon Graphics, at an insane price, and at the low-end of the market there were companies making half-assed cards that actually decelerated graphics performance. Then NVIDIA happened, and now Silicon Graphics is a has been because they didn't understand that consumers want performance at a sane price point. Today, we still have SSDs that are slower that mechanical drives at some tasks, which just boggles the mind, and on the other hand we have FusionIO, a company with technically great products that decided to try to target the consumer market by releasing a tiny 80GB drive for a jaw-dropping $1500. I mean.. seriously... what?

Back when I was a young kid first entering university, SGI came to do a sales pitch, targeted at people doing engineering or whatever. They were trying to market their "low-end" workstations with special discount "educational" pricing. At the time, I had a first-generation 3Dfx accelerator in one of the first Athlons, which cost me about $1500 total and could run circles around the SGI machine. Nonetheless, I was curious about the old-school SGI machine, so I asked for a price quote. The sales guy mumbled a lot about how it's "totally worth it", and "actually very cost effective". It took me about five minutes to extract a number. The base model, empty, with no RAM, drive, or 3D accelerator was $40K. The SSD market is exactly at the same point. I'm just waiting for a new ''NVIDIA" or "ATI" to come along, crush the competition with vastly superior products with no stupid compromises, and steal all the engineers from FusionIO and then buy the company for their IP for a bag of beans a couple of years later.

*) This really is stupid: 256GB OCZ Z-Drive p84 PCI-Express is $2420, but I can get four of these 60GB OCZ Vertex SATA at $308 each for a total of $1232, or about half. Most motherboards have 4 built-in ports with RAID capability, so I don't even need a dedicated controller!

I'm kinda fed up waiting for the SSD manufacturers to get their act together. There's just no reason for drives to be only 10-50x faster than physical drives. It should be trivial to make them many thousands of times faster.

I suspect that most drives we're seeing are too full of compromises to unlock the real potential of flash storage. Manufacturers are sticking to 'safe' markets and form factors. For example, they all seem to target the 2.5" laptop drive market, so all the SSD controllers I've seen so far are all very low power (~1W), which seriously limits their performance. Also, very few drives use PCI-e natively as a bus, most consumer PCI-e SSDs are actually four SATA SSDs attached to a generic SATA RAID card, which is just... sad. It's also telling that it's a factor of two cheaper to just go and buy four SSDs and RAID them using an off-the-shelf RAID controller! (*)

Meanwhile, FusionIO makes PCI-e cards that can do 100-200K IOPS at speeds of about 1GB/sec! Sure, they're expensive, but 90% of that is because they're a very small volume product targeted at the 'enterprise' market, which automatically inflates the price by a '0' or two. Take a look at a photo of one of their cards. The controller chip has a heat sink, because it's designed for performance, not power efficiency!

This reminiscent of the early days of the 3D accelerator market. On one side, there was the high-performing 'enterprise' series of products from Silicon Graphics, at an insane price, and at the low-end of the market there were companies making half-assed cards that actually decelerated graphics performance. Then NVIDIA happened, and now Silicon Graphics is a has been because they didn't understand that consumers want performance at a sane price point. Today, we still have SSDs that are slower that mechanical drives at some tasks, which just boggles the mind, and on the other hand we have FusionIO, a company with technically great products that decided to try to target the consumer market by releasing a tiny 80GB drive for a jaw-dropping $1500. I mean.. seriously... what?

Back when I was a young kid first entering university, SGI came to do a sales pitch, targeted at people doing engineering or whatever. They were trying to market their "low-end" workstations with special discount "educational" pricing. At the time, I had a first-generation 3Dfx accelerator in one of the first Athlons, which cost me about $1500 total and could run circles around the SGI machine. Nonetheless, I was curious about the old-school SGI machine, so I asked for a price quote. The sales guy mumbled a lot about how it's "totally worth it", and "actually very cost effective". It took me about five minutes to extract a number. The base model, empty, with no RAM, drive, or 3D accelerator was $40K. The SSD market is exactly at the same point. I'm just waiting for a new ''NVIDIA" or "ATI" to come along, crush the competition with vastly superior products with no stupid compromises, and steal all the engineers from FusionIO and then buy the company for their IP for a bag of beans a couple of years later.

*) This really is stupid: 256GB OCZ Z-Drive p84 PCI-Express is $2420, but I can get four of these 60GB OCZ Vertex SATA at $308 each for a total of $1232, or about half. Most motherboards have 4 built-in ports with RAID capability, so I don't even need a dedicated controller!

I'm also a sysadmin at a social networking site with highly dynamic content. Unfortunately, Varnish won't help much because over 90% of the cumulative CPU time is ate up by the MySQL server.

What's your read/write ratio? memcached has helped us a lot with MySQL load for read extensive databases. IODrives has done the same for write extensive workloads. Don't be scared by the price tag, one server with an ioDrive can easily replace 5 servers without ioDrives.

This sounds more like a server refresh problem than a capacity shrinkage problem. I believe that if you have a modern server capable of PCIe and your capacity needs are limited you can migrate your Clariion array to something like this and net more performance.

And if you ain't got the wherewithal to do that, how important is your data anyway?

Lost mods, but had to point this out.

$895 is less than $3500.

Rather than costing 10x as much for 2-4x the performance, it costs 2.5x as much for 2-4x the performance. Plus it's enterprise-grade, so it has a leet wear-levelling algorithm. TRIM? Doesn't need it; the controller handles all that stuff, and is OS agnostic.

After factoring that in, what you should be saying is... why bother with an Intel drive when the ioDrive is so much better value?

But I laugh at all of you. I got 1.5TB for $80. :P

"Meh, even the most expensive RAID cards loaded up with tons of RAM aren't as fast as a couple of Intel SSD's right now, so why bother with the expense?"

Clearly you aren't familiar with ioDrive

The performance degradation in the Intel X-25 is not because of a "firmware bug". All SSD's will suffer performance degradation whether or not their writing/wear leveling algorithms have been updated via firmware.

1) As ShadowRangerRIT pointed out, it is a bug.
2) These don't suffer performance degradation, so your "all" comment is 100% incorrect.

However, if you want to apply that statement to all crappy consumer SSDs powered by Intel or JMicron controllers, then I will happily submit defeat.

http://www.fusionio.com/PDFs/Data_Sheet_ioDrive_2.pdf

This company makes pretty much the fastest device you can cram into a standard PC without messing with fibre and iSCSI.

Fusion-IO has PCI-e cards that they claim sustain 570MB/s up to 1.5GBps

Can't boot from them, but there you go.

Fusion-IO has PCI-e cards that they claim sustain 570MB/s up to 1.5GBps

Can't boot from them, but there you go.

Until recently to get decent performance in a reasonable size you needed a huge SAN with hundreds of spindles. Now that you can get stuff like The OCZ Z-Drive, the PhotoFast G-Monster and of course the Fusion-IO IODrive Duo that's not really necessary unless you also need >6TB. The 50 microsecond latency is just bonus.

And oh, joy, there will be more. The SAN vendors who are betting their next year's revenue on those $million+ performance SAN's better get a plan B, and quick.

SSD storage, and rotation-free storage in general. It is not living up to expectations or promises, never mind the crystal storage methods mentioned almost a decade ago that got some really nice density.

OCZ Z-Drive, Photofast G-Monster, Fusion-IO ioDriveDuo. Density and performance are doubling every nine months at the same time price is falling by half. What's not to love?

"$-for-$ there is no contest. HDDs win."

It's really not that simple. I can deliver 1,000,000 IOPS with an HP DL785 and 6 ioDrives. It would take approximately 4,000 2.5" 15K RPM SAS drives to deliver the same number of IOPS. And that's in RAID0. In RAID10 it would take 8,000 drives. Do you have any idea how much 8,000 drives would cost? Let's go with one the smallest commercially available enterprise 2.5" SAS drive, HP 36GB 2.5" SAS at $350. Even in RAID0 with 4,000 drives you're talking $1.4 million JUST IN DRIVES! In RAID10 we're talking about $2.8 million. You can get 6 640GB ioDrive DUO's for $60,000 and put them in a single $30,000 server, and deliver those IOPS out of the PCI-e BUS (32GB/s IIRC, thats a big B, as in bytes).

Now let's talk enclosures, how about an HP MSA50 with 10 drives per U. Now we need 400U, or 9 and a half racks worth of disks. Now we need controllers, cabling, power and cooling.

Now, obviously we're talking 1-2TB worth of ioDrive storage vs 144TB worth of spinning disk storage. Of course we're also talking about $4 million vs $100k, even before power/cooling.

Point being, workload is a real consideration here.

Anyone know the real-world performance numbers of IoDrives?

I realize most SSDs perform worse than advertised, but they have quite a bit going against them.

Lots of them are MLC; with sustained write workloads they run out of erased clusters. Suddenly their write speed drops to half.

Most SSDs are also limited by the SATA controller, which is a massive limiting factor in IOps and latency.

And the SSDs themselves have shoddy controllers with hardly any cache, that are too slow to maintain high performance under demanding erase-rewrite workloads.

But the IoDrive's specs are so absurdly high, that even if they aren't real-world, it's surely faster than a meagre 350 IOps, or even a meagre 20,000 IOps. They benchmark at almost 200,000 IOps, so I'd assume real-world performance would be at least a third to a half of that.

Using six of them to get over a million IOps seems to indicate they scale well, and real-world performance may actually be pretty close to the benchmarks.

What you have to remember is some companies like to push technology. Others will try to sell you crap better than your current crap, but crappy enough that in a year they can put out better crap for you to upgrade to.

I could, indeed, write a piece of code that would wear out parts of your SSD within days.

Could you? Would you fill the disk almost completely full, and then write and delete the last block over and over with random data 100,000 times per second?

Gee, if they had though of that they might have done something really clever like include a RAM cache and a thousand extra blocks you can't see, and happily report the block written and deleted when it really wasn't, or actually write it to a different physical block each time. They might have had a stroke of genius and included logic to move least-written data to the heavily used blocks and let you bang on fresh ones now and then. It would take a real men of genius engineer to predict this pathological case and include a special purpose computer onboard to deal with it. At least it would if the engineers didn't read slashdot where we've discussed these problems to death for years and years.

They would have done it transparently in the device logic without even telling you because the device is solid state and every bit is as close as every other bit so latency is not a problem. But no, if they were that clever they would have also included some spare bytes in every block and a map so that if a bit in the middle went bad it wouldn't knock out the whole block and some sort of error detection and correction mechanism. It a fit of brilliance they might even have planned for a heavily worn block with too many burned out bits to borrow unused spare bits from another block. Gee, if they were practically omniscient they might have included programmable firmware in case they needed to push out a cure for pathological case they hadn't considered yet.

That is, if they were clever (pdf) that way (pdf).

And if you're trying that hard to break it, a spinning disk won't hold up long at all either.

The expected lifetime on the Intel X25-e is about 24 years in an enterprise server. The products of the company in TFA likewise. Use of SLC, sparing, internal error detection and correction, wear levelling and virtual block addressing add up to devices that are not only ridiculously fast - they also last a long time and degrade gracefully (pdf).

Both the Intel SSDs and the IODrive are internally massively parallel.

The startup is on SSDs only. Look up their products page. products

So, you're in the wrong ring, though the same building.

If its life depended on it, you might as well go to their site to learn what they do: http://www.fusionio.com/Products.aspx

"The module slides into certain slots inside servers. That gives the main computing chip quick access to data stored on the flash chips." I know that a general publication has to avoid jargon; but that sort of vague-but-vaguely-specific circumlocution looks like the server architecture equivalent of an awkward sex-ed class.

http://www.fusionio.com/Products.aspx

They have a test sheet there...

Their #s quoted are for 1KB and 512KB, but they show 140K IOPS random read @ 1KB...

We tested an array of 5 of the 320GB cards and tested with MS's SQLIO app and pulled some fantastic #s...

This is still just RAM pretending to be a hard drive, SATA in this case...

Let me know when they produce a NAND based flash memory card that plugs directly into the PCI bus for ridiculous IOPS...

Oh, wait... they've already done that... it's called FusionIO...

FusionIO is the only real SSD solution right now if you need to push > 80k IOPS.

These aren't the cheapest things out there, but if you run large databases and need to get your database to actually be per formant... an array of these is the way to go.

A bunch of ram pig-tailed off of SATA is just horrifically slow in comparison...

-- Dave

http://www.fusionio.com/ - why are we not hearing so much of them any more, or is there some other reason why nobody seems to mention them?

Check out http://www.fusionio.com./ You can buy something like Micron's future product today from them, and it's available today. Albeit of course there is the little matter of price.

C//

Drives using interleaved storage methodology have already been available as PCI-e x4 cards for some time now; the FusionIO ioDrive can saturate most PCI busses, and has been available for more than a year. It also blows the doors off of these claimed specs. If I remember correctly, the ioDrive is 64 devices in parallel.

So yeah, way to go playing halfway catchup, Samsung.

Depends on the SSD, the database, and the usage pattern. I wouldn't want to run a frequently-updated database on MLC flash, that's for sure. But at work we just put out production database on a Fusion-io ioDrive, which quotes a 24-year lifespan with 5TB of writes a day. The performance is amazing. Of course, now everything is CPU-bound...

End of 2010 at the very latest, and all 15K type drives are completely doomed. Go do a calculation on the rate of price change, purely on a $/GB basis, between flash and enterprise drives, take the current price points, and see when they intersect.

This is without taking into account the phenomenal pace at which flash is increasing in performance. For a glimpse of things to come, look here:

http://www.fusionio.com/

700MB/s sustained read, 600MB/s sustained write.

C//

Pardon me, but it is "blowing down the doors" (and the house too) in some tests, like this one.

If you're an enterprise that can turn IOPS into profits, you can do much better than these Intel SSD's.

IODrive for example:
100.000 IOPS, 700MB/s read, 600MB/s write: http://www.fusionio.com/Products.aspx - you get a 80GB disk for about $4.500.

- another option is to run your database/whatever entirely in ram.

One thing that seems to be true is that storage is getting cheaper and bigger all the time.

How about faster? That is my desire, and hardware SAS and SATA raid arrays are just not fast enough for what I am interested in. I saw a new device recently that was mentioned here, the Fusion IO, that is 1000 times faster, but is cost prohibitive yet, and small in size still, like 320gb is as large as offered. At $30/GB it has a long way to go before it is really mainstream and I don't see that happening for quite some time since they have their production sold out for months on end. At least that was what I was told after contacting their sales department. It is supported in the CentOS 5.2 kernel though, so that was a major plus to me, and no Windows support yet either, another major plus to me.

SSD hasn't changed the whole seek/read/write paradigm because the current batch are tailored to be retrofit to existing HBA/RAID storage systems.
Once the IO drive is here we'll have a true random access to nonvolatile storage with latency measured in nanoseconds.

I still haven't seen anywhere to buy it or heard anyone getting their hands on an ioDrive from FusionIO, but they are complaining about too high demand in Q2 (they started shipping it to big screaming customers the 7th of April).

But they have done it right: Straight into the PCIe and it comes with a Linux-driver. I just want to be sure about that it will do wonders for my databases before buying one. It's expensive, but compared to getting a machine with 60 GByte RAM or a SAN it's a bargain. And it's faster than any RAID-card, it's said, and you can RAID them if that isn't fast enough for you.

You can get SSD's with better performance than mechanical drives right now. For an extreme case, go check out http://www.fusionio.com/

you had a typo in that url, an extra period. and those devices are meant for rack mounted server boards with 4 4x PCIe slots available, although they are a low profile board, so if they give a low profile backplane as well, then you can get them in a rack mount server. (not sure if 'low profile' means 1u or 2u i am not a sysadmin)

http://www.fusionio.com/

http://www.fusionio.com./ These products can be ordered now, although it will be more than two months for delivery (they is intense demand).