Al Gore pumps $12m into cheapo TLC flash upstart
Greenbacks for Greenbytes
GreenBytes announced a good but unexceptional Solidarity all-flash array in February. It has now morphed it into what we believe may be the first enterprise TLC flash array on the market. And it has an astounding price/performance: a single VM instance costs $12.
Investors have been so impressed by the specs that Generation Investment Management, a fund founded by former US vice pres Al Gore, has handed it $12m in a B-round, with participation from Battery Ventures and GreenBytes' management.
The iSCI-accessed, and ZFS software-based Solidarity array has three tiers of solid state storage which vary by performance, capacity and endurance. The whole system comes in a 3U enclosure containing two canisters which have 6-core Xeons inside them and provide high-availability at the controller level.
The top storage layer consists of a 48GB cache. There are two STEC ZeusRAM drives front-ending a middle layer of two STEC MLC (2 bits per NAND cell) drives which have a 30,000 P/E (Program/Erase) cycle rating. After that they wear out. The middle layer is for metadata and intermediate data storage. The top layer is for write coalescing, grouping random writes into a more sequential state and write logging. Incoming writes go into this RAM and can stay there for five to 10 seconds while they are coalesced and then sent to the third layer.
The third layer is the surprise. It is termed cMLC (consumer MLC) but is actually TLC (three layer cell) flash with 3 bits per cell and a 5,000 P/E cycle rating, which is used for capacity storage. That rating is another surprise, as raw TLC is generally supposed to have around 1,250 P/E cycles in 39-30nm flash. The 2.5-inch SSD supplier – undisclosed by the way, as is the TLC NAND chip supplier – has done good things with its controller in the wear-levelling, write reduction and provisioning areas to bump the raw number up. The capacity of these TLC SSDs is up to 2TB, that's twice as much as the maximum 1TB you can get from a 2-platter, 2-5-inch hard disk drive.
As far as we know this is the first TLC NAND array for business use in the industry. The conservative GreenBytes CEO, Bob Petrocelli, agreed that was "possible" but wouldn't go further.
ZFS dedupe replaced
Solidarity's software carries out compression and in-line deduplication and that boosts the raw capacity three to five times; it varies with the application, so that a 13.5TB raw system can hold up to 67.5TB of data in its 3U enclosure. GreenBytes wrote its own deduplication software from the ground up – Petrocelli says ZFS dedupe sucks – and puts the dedupe metadata into flash separate from the data.
It paid a lot of attention to what happens when deduplicated data is deleted and found a way, after a lot of tweaking and tuning, to do this quite quickly without having cascading changes that slaughter performance. Dedupe is always going to be CPU-intensive but it can be done, using flash memory, in ways that don't affect overall performance relative to disk drive arrays very much at all.
GreenBytes says this about its software: "A specialised data structure and search algorithm enables a proprietary deduplication capability that delivers the same high performance with hundreds of terabytes in a datastore as it would with just a few megabytes."
It claims its customers can get a 92 per cent data reduction rate with VDI. The dedupe is carried out initially by hashing blocks and then looking for a 99.9 per cent partial key match, repeating that if a match is not found at first, and repeating it again to eventually find a match in the slowest case.
Petrocelli says the 5,000 P/E cycle TLC drive's arrival is significant: "I think that with the 5K P/E flash drive we'll see a transition in the market more quickly than we imagine ... We are seeing a 50 per cent reduction in the cost of flash. You will see "me-toos" from other suppliers. The cost/GB probably in one year will be less that $2/GB. You can pretty much forget about 15K disk drives altogether."
VDI instance pricing
He said flash removes a storage I/O bottleneck that has slowed servers down as they have gained faster multi-core and multi-threaded processors. The CEO said GreenBytes has been working with a Netherlands customer and found a single Solidarity array, with its 140,000 IOPS (8K block size), could support 10,000 VDI users.
Dividing that box's list price of $119,000 by 10,000 gets us $11.90 per VDI instance, a potentially killer number. According to the numbers we have:
- The hybrid flash/disk Tintri array's cost/VM is $23;
- Dell's EqualLogic PS6010V is $338;
- HP's all-flash P4000 P4900 is $233; and
- a NetApp solution sets you back $51/VDI (PDF).
The Tintri cost/VM simply slays the bang-for-buck of Dell and HP, and is half NetApp's cost. The GreenBytes cost/VM is almost half the Tintri cost, a quarter of NetApp's, and laughably less than the EqualLogic and LeftHand iSCSI array costs.
But that's the magic of TLC flash pricing. We're told 10K SAS disk drives are at a $1/GB cost or less. Deduped and compressed TLC flash is at $0.80/GB. The dedupe and compression sacrifices some IOPS for capacity but the system is still way faster than spinning disk.
The Netherlands customer will roll out 750,000 virtual desktops in 12 months, Petrocelli says, adding: "You couldn't economically provision this product with 15K disks."
Competing vendors will point out that GreenBytes being a startup. But its core software is pretty mature: it has been developing its deduplication technology since 2007 and delivering it in products such as the 2009 GB-X deduplicating appliances and other hybrid flash and disk arrays.
GreenBytes noticed that, with is deduplicating systems, there was a significant reduction in the numbers of writes to the storage pool relative to the number of writes coming in to the system overall. Lots of updates are made to transaction logs for example but many of these can be deduplicated away. Petrocelli said; "We realised this would be a great application for lower-cost flash [and] grew into it as we separated metadata from data."
Now GreenBytes has an all-flash arrays and Petrocelli reckons hybrid arrays are like; "Flash bolted on to a dinosaur to make a mammal when the mammal is the all-flash array."
We hear that the Solidarity array will gain a Fibre Channel interface quite soon. It almost goes without saying that a Solidarity array takes up much less space, draws less power and needs less cooling that a traditional disk drive array; the GreenBytes array is green. Petrocelli says it reduces the carbon footprint of information.
The $12m B-round funding brings total funding for the start-up to $30m. The cash will be used to expand and accelerate global sales, marketing and channel development efforts, and for research and development. GreenBytes is focused on the small and medium business market, its CEO maintains, and is not in competition with Violin Memory and other high-performance flash array suppliers, he says.
If this TLC flash-based array is as good in real life as GreenBytes' specs say it is, then a new front has been opened in the flash storage war against disk drive arrays. The key is the TLC's endurance and GreenBytes' technology's ability to de-amplify writes so as to make the best use of the TLC stuff. If it has pulled off that technological trick and the TLC SSDs perform as well as their supplier – whoever it is – says they do, then we have a potentially knock-out flash array on our hands.
TLC use will spread. A NetApp spokesperson said: "We use SLC today. We are testing eMLC and TLC and expect to be able to support those within the Data ONTAP architecture as customer demand requires."
List prices for the TLC Solidarity array are $59,000 for a 3.5TB raw configuration, $73,000 for 5.6TB and $119,000 for 13.5TB. ®
Sponsored: Hyper-scale data management