When am I getting my VVOLs mang? I'm tired of these RDMs.
Fresh from a big product and business model launch, Pure Storage was undestandably reticent about future technologies at a briefing in Mountain View. However, interesting tidbits did drop from speakers' lips. Co-founder John "Coz" Colgrove said: "Our goal is to make all disk storage extinct." Dressed in his signature long t- …
Right now, the Raw $/GB of desktop for consumer-grade, 960GB MLC SSDs is about $0.50/GB, which is about the same as OEM prices for 900GB 10K RPM SAS drives.
These SSDs are not enterprise grade, they target the home PC gamer market. They probably have MTBFs too high for use in storage arrays. But by next year, enterprise grade versions of these will be readily available, and likely at a similar $/GB as this years PC SSDs.
3D MLC NAND is starting to come out now, putting price pressure on standard MLC. Also, standard MLC has benefited from die shrinks and the manufacturing learning curve. The result is all MLC flash is cheaper this year. By next year, it will be cheaper still.
Dedupe will still matter, because it will provide competitive advantage. But it will not be needed for NAND flash to match SAS HDD.
Oh dear lord. You do know that 'Enterprise' grade SSD's are simply the same commercial SSD drives but with more capacity hidden away for internal processes like garbage collection?
Made in the same fab as commercial SSD drives
Made with exactly the same controllers
Ships with exactly the same firmware
The notion that somehow there is a magical fab churning out 'better' SSD drives is not true - the only difference is the overprovisioning and then whatever the vendor chooses to do with them.
This post has been deleted by its author
So Flash is here for a while, not gonna argue that. Its also going to continue to erode disk. Is it ever going to fully replace it, or even come close? I seriously doubt it. That said, John has every reason to peddle that thought as its good for his group (FYI, I dont fault him in the least for that).
The short version is that theres just not enough production capacity. Even if they magically could massively up the production capabilities, flash foundries are CRAZY expensive. If theres enough supply to truly replace disk, its price has been massively eroded which reduces the return on investment, which the foundries have a strong incentive to avoid. Tack on the weirdo-issues with increasing densities (read disturb, reduced write capacities, etc) and its suddenly less rosy picture.
My hypothesis (worth exactly the zero you payed for it) is that it'll continue to erode disk slowly, but disk will just move down the hierarchy, rather than ever be completely replaced. Consider how people have been screaming how tape is dead for many years yet its still there in pretty significant quantities.....
This post has been deleted by its author
Although I agree with the vision that SSD will replace spinning rusty stuff, we're not there now, nor will be in the near future 3 years based on what's on the horizon.
I work in the storage consulting business - have experience with multiple brands - and have yet to see an SSD array that hits a price point close to spinning or hybrid options.
Some of the technology developments - compression, deduplication, tiered MLC/SLC, etc - alll accomplish the same thing: lowering the cost per GB - and have allowed the price points to get closer to trad storage for SSD arrays, but the ExtremIO, PURE and other SSD options are still being sold (and apparently bought) at a significant premium to market.
Sure, some customers need the outright performance, as well as some customers are wanting to buy into the leading edge technology, but many customers are wasting money on technology they neither need from a performance perspective, nor need from a growth point of view.
Not every application needs a Ferrari: most applications work very well with a 4-door saloon or a bus.
Deduplication and compression also apply to HDD storage as well, an oft-overlooked fact, especially when looking at cold storage, which is the vast majority of data. Sure, disks cannot absorb the performance hit quite as easily, but you are burning CPU cycles for all that dedupe and compression anyway no matter the medium, and for cold data the performance overhead is tenable.
Yes and no. Dedupe and compression ala Pure could work on something like a big old array of SATA drives but I think you underestimate the performance hit. It would essentially relegate the disk to back-up/archive only and even then I question how effectively it could do a restore. As data becomes deduped it changes it's characteristics from sequential to random. Sure you have the possibility of retrieving some of those blocks from cache but unless it is very highly deduped you'll not make up the meta-data overhead and random access penalty. Depending on how compression is implemented there can sometimes be a performance bump but more often then not it consumes a little performance. The two real reasons dedupe and compression work so effectively on flash is that first it has performance to burn. Until recently SSDs were smaller in capacity then there SAS counterparts requiring more drives and the drives themselves are an order of magnitude better, more so in a very random environment. You could give up the same number of IOPs you're using on SAS for backend processing and barely notice. Second the random characteristics don't phase SSDs at all. Two reads from entirely different parts of an SSD to construct a block, no problem, two reads from different parts of a disk, have fun, seek time penalties are going to kill performance.
The reason disk never replaced tape was due to the high ongoing cost to keep it alive in a datacenter. Disks require a lot of power, cooling, and rack space, and their footprint only grows over time, all while dealing with failures of the mechanical bits which drive up operational costs to replace them continually. Flash technology doesn't have these same limitations. Flash will absolutely replace disk and tape.
For those that don't believe this consider it took 50 years for the HDD to get to 1TB. Samsung is shipping an 8TB SSD later this year. In order for the HDD manufacturers to increase drive density they need to retrofit their assembly line for HAMR, and even that technology hasn't won out yet. HGST was doing helium filled, SMR is here, and even bit patterned media is being toyed with. Either way, the cost is substantial to retrofit, and the big HDD manufacturers have acquired NAND media makers over the past few years. The writing is on the wall. Invest in HDD tech, or invest in flash? The market is speaking now, and if you listen, you will hear it too; disk is dead.
Biting the hand that feeds IT © 1998–2022