Slashdot Log In
RAM Disk Puts New Spin On the SSD
Posted by
samzenpus
on Thu Jan 22, 2009 07:57 AM
from the old-disk-new-tricks dept.
from the old-disk-new-tricks dept.
theraindog writes "Although the solid-state storage market is currently dominated by flash-based devices, you can also build an SSD out of standard system memory modules. Hardware-based RAM disks tend to be prohibitively expensive, but ACard has built an affordable one that supports up to 64GB of standard DDR2 memory and features dual Serial ATA ports to improve performance with RAID configurations. And it's driver-free and OS-independent, too. The Tech Report's in-depth review of the ANS-9010 RAM disk pits it against the fastest SSDs around and nicely illustrates the drive's staggering performance potential with multitasking and multi-user loads. However, it also highlights the device's shortcomings, including the fact that SSDs are more practical for most applications."
Related Stories
This discussion has been archived.
No new comments can be posted.
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
Full
Abbreviated
Hidden
Loading... please wait.
What I learned from the article (Score:5, Insightful)
Re:What I learned from the article (Score:4, Insightful)
I'd bother, because most of *my* tasks are disk I/O bound.
Parent
Re:What I learned from the article (Score:5, Insightful)
But are you really getting your money's worth from this device?
DDR2 is an order of magnitude faster than SATA. Looking at their numbers, the internal controller is limited to about 400 MB/sec. That is pretty mediocre.
Parent
Re:What I learned from the article (Score:5, Interesting)
What if the ramdisk function was moved into the motherboard chipset? This would achieve 2 things:
1. It would dramatically cut the cost of a ramdisk. I.e. The cost of the entire motherboard might go up by $5 or so.
2. It would eliminate that SATA bottleneck, allowing ramdisk to run at full RAM speed.
If you then figure out a way to have this data loaded to the ramdisk from a hard drive at poweron (or get realy clever and mount a flash chip on or near each DIMM which takes a backup of that DIMM, just before powerdown.
Parent
Re:What I learned from the article (Score:4, Insightful)
This brings up an interesting idea. What if the ramdisk function was moved into the motherboard chipset?
OMFG! That's an AMAZING idea! This could dramatically change computing as we know it! The implications of this are, eh, well....
.... quite well understood. Somebody thought of this many years ago. Many, many, many years ago. It's called a (ahem) "ram disk" and uses system memory as if it were a drive with a software driver. Here's a howto for Linux [vanemery.com] - I did something similar with so-called "high memory" on a 80286 with DOS 3.x and ramdrive.sys [geocities.com] - that 384k ram disk was small, but //FAST//!!!
Sorry to break the news to you.
Parent
Re:What I learned from the article (Score:4, Interesting)
That's not what I want. I don't want to have to buy the fastest, most expensive RAM available just to use as a RAM disk. I'd prefer cheaper RAM, maybe two or three generations old, that I can get in massive quantities.
Except that you can't get two or three gen old RAM in massive quantities. At least, not for as cheap as the new stuff. See for yourself: 1 GB of PC 133 RAM is more expensive than 1 GB of DDR or DDR2 RAM [pricewatch.com]. There's a very short window of "cheaper" just behind the bleeding edge that's cheaper than the very latest (DDR3) but new motherboards support this type of RAM too, negating the "two or three gen old" situation that you state.
Most people think that the older the technology, the cheaper it gets. But this is only true for a very small time window, at which point the older technology gets phased out (not profitable, anymore) at which point it becomes a "niche" marketplace with very low volume and very high prices.
Example: A PC-133 RAM stick needed to keep a $12,000 vertical-market weaving loom operational, where the cost of the additional RAM pales compared to the cost of the entire integrated system. If you need that extra 512k of RAM in your $12,000 loom, paying $100 for it isn't such a bad deal.
But there aren't many people stuck with $12,000 looms, so the price of the older technology skyrockets until it's simply not available anymore. (Ever try buying a *NEW* Cx 6x86 processor in the last year or so?)
Parent
Re: (Score:3)
This brings up an interesting idea.
The real question that is not being answered here is why does a 32 GB SD card cost $25 but a 64 GB SATA hard drive cost $800? Why can't the technology that makes SD cards so cheap make cheap SATA hard drives as well?
If I'm missing something obvious and this sounds like a troll, then please RTFM me with a link, because I'd really like to know.
Re: (Score:3, Informative)
Re:What I learned from the article (Score:4, Interesting)
The real question that is not being answered here is why does a 32 GB SD card cost $25 but a 64 GB SATA hard drive cost $800?
A 32GB SDHC card right now on Newegg (in-stock) is a minimum of $72. I don't know where you got the $25 number (sure, in another year it will be that cheap). As another poster mentioned, Newegg has 32GB SSDs available for the same price range.
Why can't the technology that makes SD cards so cheap make cheap SATA hard drives as well?
It already has. The first SSDs on the market used single-level cell (SLC) flash [wikipedia.org], while the inexpensive SD cards and mp3 players you see everywhere use multi-level cell (MLC) flash [wikipedia.org]. The difference is how densely you can pack the data, and it makes a huge difference in price.
To put it simply: SLC flash is faster, lower-power, and more reliable than MLC flash, but also more expensive (at same capacity) than MLC flash.
The reason the first SSDs used SLC flash is because new technologies have to convince people to take the plunge: people/companies are usually willing to pay significantly more for something that is much faster and more reliable. Early adpoters might have given SSDs the cold shoulder if the first wave of drives reduced capacity and performance in-order to be more cost-competitive with existing storage.
Now that SSDs are firmly off-the-ground, manufacturers are offering all sorts of devices, including cut-rate drives using MLC flash, so the prices at the low-end have dropped like a rock.
Parent
Re:What I learned from the article (Score:4, Insightful)
You can get SSD cards with a PCI-e interface that hit 800MB/sec. Why RAM disk manufacturers stick to SATA I don't know.
PCI-e even has from standby power available.
Parent
Re: (Score:3, Interesting)
Bingo.
That's why I'm holding out for a FusionIO. Their cards go through PCIe4x - not SATA. I want 600MB/sec reads/writes! 60k ops/sec, and cheaper than this thing!
I just hope their consumer grade products perform as well as their enterprise ones. Apparently tech report will be reviewing them as soon as they can get their hands on them.
Re: (Score:3, Interesting)
Mac OS7 had a way of creating a RAM disk from installed RAM. Was sweet way to run Photoshop at top speed back then, along with the dedicated Photoshop video card. Such specialization back then.
Re: (Score:3, Interesting)
Re:What I learned from the article (Score:5, Insightful)
Why would you spend your money on this device instead of just buying the equivalent amount of RAM and putting it on the motherboard where the processor can access it directly? Even if you had to upgrade to a more expensive motherboard you'd still get way better price-performance by doing that, rather than crippling the RAM by putting it on the other side of the SATA bottleneck.
If you insist on having a 'disk' you can save files to, well, all OSes support the idea of a RAM disk...
Parent
Re: (Score:3, Informative)
it needs about 20 minutes to transfer 16GB to backup card
That's 16 megabytes per second - if I had to guess, the bottleneck is the CF transfer rate and has nothing to do with the rest of the device.
Re:What I learned from the article (Score:4, Informative)
You're confusing bits and bytes.
Parent
Why are these always so expensive? (Score:3, Interesting)
Without RAM, this costs $380 which is probably more than double the RAM itself if you don't use anything to extravagant. I know other companies offered these in the past, with the similiar high price, always to act as a harddrive with a battery for backup. It was always easy in linux to make a portion of your memory act as a ramdisk, however many motherboards often didnt enough ram slots to make it appealing to split memory up like that.
I wonder if a company like Apple can instead, on its laptops for instance, just move to SSD for its laptops since they are becoming seemingly cheap, exclusively (for OEMs) license a technology like MFT, and get a real speed edge on other makers. I think it would make more sense than a ramdisk where the bandwidth of ram vs the hard drive channel seems overkill.
Re:Why are these always so expensive? (Score:4, Insightful)
Parent
Re: (Score:3, Insightful)
Volume. Will always be a niche product, so they have to sell it at a high price. Now, if Dell or somebody did a buyback scheme of their old PCs and recycled the memory in some kind of cheaper version of this box...
Re:Why are these always so expensive? (Score:4, Informative)
Intel's new SSD's have MFT-style stuff on the controllers; they have very fast random writes [blogspot.com].
In contrast, the Intel SSD does about 8,500 4kB random writes per second.
Parent
No ECC... (Score:4, Interesting)
so, this is just as worthless as Gigabyte's i-RAM.
Re: (Score:3, Informative)
I've got one. Registered ECC is not supported. Unregistered ECC is supported. I saw no real performance decrease in simulated vs real ECC RAM. The SATA interface seemed to be a much greater bottle neck.
ahh 1993 was a good year (Score:2, Redundant)
Re: (Score:3, Informative)
These days you can just use iSCSI to any free Unix-like and export a memory-backed virtual disk. It's also a nice way to use one machine's memory as swapspace for another, and with a fast network link it's like having more RAM in the client machine.
Great for swap and /tmp (Score:2)
This seems like it would be an excellent solution for a swap drive and for a c:\temp or /tmp directory.Fair enough as swap *nix OSes will reap little benefit, but Windows seems to hit swap no matter how much RAM you have so there should be some significant performance gains there.
Re:Great for swap and /tmp (Score:5, Insightful)
Parent
Re: (Score:2)
Re: (Score:2)
I agree. There was a time when it made sense to use an i-RAM to get around the 4GB limitation, but with 64 bit readily available I wouldn't begin to consider this unless you need >16GB RAM. Even then you're likely to find server motherboards with more slots and support for higher densities, not to mention more sockets - I would imagine that that most applications which need this amount of RAM also could do with more processors.
Re:Great for swap and /tmp (Score:4, Informative)
My apologies - forgot I wasn't supposed to RTFA.
Parent
Re:Great for swap and /tmp (Score:5, Informative)
If you compare this thing to just putting the RAM in your PC there are NO upsides.
Ok...
1. Find me a motherboard that has 8 RAM slots that doesn't require expensive ECC and/or Registered memory
2. Find me a computer that can boot from it's own RAM drive.
3. Find me a computer that can use a RAM drive that can be persistent through reboots without having to save the contents to something else.
I have several of these, and I run a power cord that is normally used for one of those SATA/IDE to USB kits in the back of my computer to power my box.
You don't think about all of the uses that this thing offers.
Parent
Re: (Score:3, Interesting)
The whole thing is pointless - why not just put 64GBs of ram in your PC and let it fill it up with disk cache. This makes no sense. If you compare this thing to just putting the RAM in your PC there are NO upsides. The data is vulnerable, it's massively expensive and an inefficient use of the RAM modules. Madness.
Well, from everything on this product that slashdot has mentioned, just sticking RAM on a motherboard would be a better solution. It's not always the best though.
I've wanted one of these things for
Re: (Score:3, Funny)
Re: (Score:3, Informative)
IA32 can address up to 64GB with PAE.
In summary (Score:5, Informative)
Skimming the article, I'd summarize as follows:
Real world performance not radically better than fast traditional HDs or SSD solutions, and you can't power off your PC for the night. (Unless you backup to flash every night.)
I'd say this is a niche product, but could be a very good one for a chosen few applications.
Re: (Score:2)
what's the difference from a tmpfs then?
Re: (Score:2)
what's the difference from a tmpfs then?
There's a battery that lasts a few hours or something, but not through the night. RTFA for proper details. ;)
I RTFA and now speculate (Score:5, Interesting)
Re: (Score:3, Informative)
I've got one. In practice, it is. You cut power to the device, it will start making a backup using its internal battery - which lasts 3-4 hours. This is not dependent on you pushing the button to do a manual backup of the current drive image.
Pagefiles, watch out (Score:2)
Yay, a RAM HD! I'd like to see the pagefile dig into this - Microsoft must be foaming at the mouth. Sorry if that seems like trolling, but I've had it up to here with the constant and painful HD thrashing that Windows always seems to enjoy doing (and probably their less than perfect implementation of it).
Re: (Score:2)
Yay, a RAM HD! I'd like to see the pagefile dig into this - Microsoft must be foaming at the mouth. Sorry if that seems like trolling, but I've had it up to here with the constant and painful HD thrashing that Windows always seems to enjoy doing (and probably their less than perfect implementation of it).
You'll be better off sticking all that RAM in your computer than in a $380 box. (Or, optionally, spend the $380 to upgrade your mainboard to one that can hold all the RAM).
Not that impressive (Score:2, Informative)
Nothing new (Score:3, Interesting)
I have a high-load mail server that uses a 2G RAM disk (a Curtis Nitro!Xe [curtisssd.com]) for the queue. It looks like a normal 3.5"/1" high SCSI drive with a SCA hot-swap connector. It was made before high-density CF cards, so it has a 2.5" notebook hard drive inside for storage after shutdown (it has a battery to start the drive, dump the RAM, and shut down). We've had this in service for almost 5 years, and it has really made a difference.
The point to a RAM disk is not necessarily bulk data throughput, but I/O operations per second. Mechanical drives are limited to 100-200 random IOPS or less, while the RAM disk can easily hit 100,000.
Good idea (Score:5, Interesting)
My university used RAM disks back in the day - it was the only way to get decent performance on older machines. The computers didn't even have hard disks in. My brother (who went to the same university) has a story where he sped up his large FORTRAN compiles by a factor of 10 just by working out how to use the RAMDisk (which was only ever used by the PXE-style boot procedure and then hidden from the OS) for his own purposes and people couldn't work out how he was doing it because he still took stuff home and brought it in on floppy. This is a nice hark back to those times.
The killer, however, is the price... the price of a PC, basically, before you add the RAM. If you're REALLY serious, you'll have machines that can just take the extra RAM directly and do this in software. If you're not willing to pay that much, well, nothing will work for you but a few bits of extra RAM and a fast SSD for the same price won't go amiss. However, if you occupy the middle ground... this still doesn't seem worth the effort. It'd be cheaper to just buy an SSD, some extra RAM for cache and maybe even a cheap PC to throw it all in (if NanoITX supported 8Gig chips, this device could almost be made obsolete overnight).
The interconnect too - yes, it emulates a SATA drive but it emulates two as well and fails to do anything significant with them. So you'd need a RAID0 setup, with independent SATA setup, and an expensive device, with lots of even more expensive RAM just to be a fraction of a second quicker than an off-the-shelf SSD in the same machine. The people for whom it's worth it won't want to be bothered with all this.
The CF Backup feature is fantastic. I love the idea. But 20 minutes is a long time to wait if the battery is only four hours worth when it's brand new (four hours? At least 24 would have been useful and given you a chance to actually do something with it). You would want to be backing up anything this thing held anyway, so you don't really gain anything because the CF is the most inconvenient backup because of its manual nature.
I can't see a situation where 64Gb of fast storage is worth that amount of money + time + hassle + 64Gb of RAM + potential firmware problems + interface cabling + ... The bottlenecks in anything serious are going to be elsewhere.
Much as I like this, I can't see the market (Score:3, Interesting)
First of all, I absolutely love these devices. It's a great idea that's been well executed, and yes, they're a niche product, but we've one or two apps that would notice the increase in speed from these, and if I had the money I'd buy a whole bunch of them to stick in our servers. ... except that you don't get 5.25" bays in an awful lot of modern rack mounted servers. Certainly none of our new kit has them; all that space is taken up with hot swap 3.5" or 2.5" drives.
And that's what kills it for me. Even if I'm looking at a new server I'd have to make sacrifices to fit one of these. My first choice for a new storage server is going to be one with 24x 2.5" drive bays. I'd have to sacrifice a full 8 drive bays to make room for one of these, and it's just not worth it. Not when I can buy an Intel SSD for the same price, loose just one bay, and have it hot swap to boot.
And even worse, there are PCIe devices just around the corner, with 3-4x the read and write speed of any SATA device. Those will drop straight into any of our current servers, no problem at all.
So unfortunately, much as I love the ANS-9010, I just can't see any reason to buy one :(
failure mode (Score:3, Informative)
How does this thing handle getting the power cord yanked in the middle of a large write operation?
Re:failure mode (Score:4, Informative)
Same as any HDD - a hard shutdown. The battery pack will then start backing up the current state of the memory to a CF card, so that when power is returned to the system you can run fsck or chkdsk. If you don't have a CF card, it will keep the RAM alive for a few hours, then all is gone if power was not restored.
Parent
Re: (Score:3, Funny)
How does this thing handle getting the power cord yanked in the middle of a large write operation?
Since you didn't RTFA, you'll be happy to know that you can keep on gloating. The thing has absolutely no backup mechanism, no battery, no ability to write to a CF card. If the power so much as winks, all of your data is garbage.
Son of iRAM (Score:5, Informative)
A few of the things it improved on the old iRAM.
*DDR2 supported ram, with 6-8 slots, taking up to 4G sticks.
*A fair sized battery.
*A CF backup slot.
*RAID friendly, multiple SATA ports on 8 slot model.
*Uses 5.25" bay rather than PCI slot.
*ECC
First off, no special device driver was needed - the drive was OS agnostic. Every mainboard and controller card I used saw it the device like any other SATA hard drive you might plug in.
The RAM slots take bog standard DDR2 RAM. The documenation mentions speeds of 400/533/667/800 are all supported. Benchmarks with 533 and 800 grade RAM produced identical benchmarks, so faster RAM does not appear to have any impact. I also mixed and matched faster and slower DDR2 modules without issue.
Just like most mainboards, the RAM needed to be installed in pairs if over one stick was used.
Unbuffered ECC or non-ECC modules are both supported. Registered RAM was not. I tried to pull eight 4GB sticks from one of my Sun boxes to give the 'full montey' test. No joy. Had to stick with the far cheaper RAM.
There was an interesting option for these who wanted to have ECC but used 'regular' non-ECC RAM. Eleven percent of the memory could be reserved for error correction. Again, all hardware based - just move a jumper. Performance metrics between ECC and 'simulated ECC' had negligible differences.
The 8 slot model has two SATA ports. By setting a jumper, you could have the entire RAM capacity as one large drive on one SATA port or split it as two independent drives. If you splid the drive you had to have an even number of RAM sticks installed. Another jumper would dumb the interface down to SATA1 speeds rather than SATA2. Never tested that....
Did test RAID-0, however. (grin) The synthetic benchmarks don't hit this device's sweet spot - database usage. Reads are fast. Writes are just about as fast. The RAID controller really makes a difference, as my 3Ware card performed significantly faster than with the mainboard based RAID. Using a EVGA 780i mainboard, it was not crushingly faster than a trio of velociraptors.
For anyone who has installed XP, you know the wait between hitting the 'workgroup' and the first reboot? Just over two minutes. By far the fastest install I've ever done. The OS also started faster than any other disk or SSD system I've used.
The CF bay was a nifty option. The question came up - what if I want to shut my machine down overnight? You can. If you have a CF card with more capacity than your RAM, it will back up the disk image automagically. You can also push a button to back up the current 'drive image' to CF, and another to restore the image. (I was able to go back and forth from Linux and Windows very easily).
Anyhow, tis a fantastic high speed scratch disk or OS disk when write speed matters. For those of us who already maxed out RAM, this covers the gap between RAM drive sharing RAM with the mainboard and fast disk.
Re: (Score:2)
The "ANS-9010 RAM disk" is the device with the shortcomings. SSDs are (in the author's opinion) more practical.
Re: (Score:2)
...Since when is being "...more practical for most application..." considered a shortcoming?
When any attempt to making use of the results of being practical is more of an exercise in frustration than maintaining the status quo.