I think what I'm most amazed by is people update facebook as often as they search google. Genuinely surprised by that.
For a non-volatile storage medium flash development sure is volatile, as attendees at a distributor conference found out. Aaron Rakers, managing director of analysts Stifel Nicolaus, attended the conference and sent out a despatch from it that contained a mouth-watering array of tit-bits. The distributor had suppliers of hybrid …
695,000 is 1 billion a day, divided out by 1440 minutes. It's a suspiciously round number which just tells me that facebook and/or google don't really know or want to share the answer.
Memristor vs. Flash
If you could choose to store valuable data on a device that used a complicated VLSI chip to encode and store your data on a large-block-addressible medium that you know was likely to start failing after a few thousand write cycles, or one that used simple addressing logic to store it as individual bits or bytes on a medium that lasts for billions of write cycles, which would you choose?
HP will win hands-down, if they can get memristor storage down to much the same price per Gbyte.
If the theory translates into practice, memristors will be the biggest revolution in computer technology since CMOS arrived.
Re: Memristor vs. Flash
On the other hand, Flash has a real-world history to go by, whereas Memristors' performance characteristics remain at this point theoretical, and even upon launch will have very short-term lab history to validate HP's longevity claims...
Re: Memristor vs. Flash
"On the other hand, Flash has a real-world history to go by"
Yes, but Flash's history isn't exactly that brilliant. In fact its bloody awful. All that fussing with wear leveling, block erasing, error correcting and all that fretting about whether it's gonna just forget what's been written is a pain in the rear. It's only plus point is its moderately useful capacity (and a little bit non-volatile). Otherwise, in absolute terms, it is fairly painful to use.
Whilst it's true us ordinary mortals haven't got our hands on a memristor to evaluate it for ourselves, HP have been publishing all sorts of promising looking papers on their performance, longevity, etc. Remember, even if HP have to back off from what they say they can do by a factor of 10 (speed, capacity, longevity, take your pick) they're still going to comprehensively obliterate Flash.
Re: Memristor vs. Flash
"Yes, but Flash's history isn't exactly that brilliant. In fact its bloody awful. All that fussing with wear leveling, block erasing, error correcting and all that fretting about whether it's gonna just forget what's been written is a pain in the rear. "
All that may be moot if the "bake" tech proves to work as well in devices as it does in the lab. At that point write cycles become effectively indefinite.
You don't scale performance linearly...
The statement that 32GB/s Fibre Channel would take "just under 4 minutes" to transfer 1TB is bogus: doubling line speed doesn't double transfer rate. Actually, the figures presented show the reality:
Quadrupling Ethernet from 10Gb/s to 40Gb/s increases performance 373%, not 400%. So using THAT metric, 32GB/s FC would take four minutes and 15 seconds, although "just over" would be accurate. "Just under" is not.
Who wants to take the bet that HP won't be shipping memristor anything (other than perhaps coffee cups) by the end of 2013?
Is it too soon to order my "I bought an array from HP and all I got was this lousy mug" mug?
I think that they've already started stamping out test devices.
If HP can pull this off then they'll clean up. They will end up owning the entire storage market. It is interesting to note that in a company that has ditched almost all its blue skies research this project has survived. Clearly they have quite a lot of faith in it. Given that it is such a game changer its worth almost anything, even the odd delay or two.
Re: Too soon?
I did (several arrays) and didn't even get a mug!
HP has been too much all over the map with their prognostications. I think they're mostly trying to hang on to their jobs. As fantastic as the technology sounds, I think it probably has a long ways to go in the lab-fabs before it makes it to production -- if that ever even happens. Just the other day Samsung said it would cost $10B to build a fab that can produce memristor memory. I can't find that link anymore, but it sure sounds like there's a lot that's not being said.
<disclosure: I am not a employee or PR person for any of the companies below>
Isn't Skyera doing today what HDS is proposing in 2015? Skyera's SEOS and Skyhawk is claiming <$1/GB effective capacity.
What I think the analysts (tech and fiscal) often skip is the fact that flash and solid state storage doesn't need to be packages in the same form factor or behind the same storage level protocols as spinning disks. They can be their own unique packaging keeping the SAN/NAS/PCIe front end.
Violin Memory pioneered this to a certain extent at the array level and FusionIO did so with others at the PCIe level. Skyera does this as well. This can work as long as they can stay ahead of the cost/GB and performance curve of SSDs or what ever the top flash vendors work out to be the spiritual follow-on to SSD (I wonder if vendors such as Micron, Intel, Samsung and Toshiba are seriously looking at this space and thinking of ways to create commoditized building blocks using a PCIe interconnect (NVM Express) to replace SSDs leaving storage vendors to create software around the areas of innovation mention below)
Vendors doing all flash arrays such as XtremeIO, Pure, Whiptail, Solid Fire, Nimbus, Tegile etc-- all using commodity SSDs are being lazy about thinking outside of the box on the hardware front. That's not a sound long term strategy as the software functionality of compression and dedupe becomes common pace.
Other areas of SW innovation include:
1. Scale-out storage ... equivalent of LVM functionality at the macro level with load balancing and multi pathing
2. OS/Application integration with functionality such as VMware VAAI/VASA etc, Microsoft VSS, Oracle and so on
3. Data protection (snapshot, replication/DR, tiering) ... something for file systems to consider (ZFS, BTRFS etc)
4. Unified storage (multiple protocols -- NAS, SAN) ... sort of commoditized by the likes of Open Solaris/FreeNAS
5. Management and reporting ... Example of Tintri comes to mind here with their VM Aware storage