Comment The best situation for an all-flash array is to be built from commodity flash components and be at a WAN arms-length from servers; that’s SolidFire CEO Dave Wright’s view.
SolidFire’s SF9605, SF4805, and SF2405 arrays run the Elements OS, now in its 8th major release. The systems have a clusterable, scale-out design, an unlimited drive wear guarantee, and the software has a focus on multi-tenancy and quality of service for cloud service providers.
We met CEO Dave Wright in SolidFire’s City of London offices, and set about finding out more about the company’s strategy.
It has adopted Samsung 3D V-NAND flash, and that enabled a big price drop when the three latest product platforms listed above came out earlier this year. Wright said 3D TLC NAND is good stuff. Everyone is working on it, and QLC (Quad Level Cell) comes out next year.
QLC has 4 bits/cell, compared to TLC’s 3 bits/cell, so a third more capacity in the same physical space. Wright thinks the quad level term is a misnomer, as the cells actually have 16 voltage levels inside them, with TLC having eight and MLC four.
Wright seems sure that QLC flash is real and has a degree of faith that Intel’s Silicon Photonics tech is going to be real too. So we could perhaps see QLC tech entering SolidFire’s engineering labs next year, if prototype products are not already there.
Sticking with COTS SSDs
The company uses 2.5-inch SSDs, bought off the shelf. Would SolidFire build its own SSDs or use NVMe? He said: “We’re certainly not going to engineer our own hardware. That’s a dying idea.”
“NVMe flash is more expensive. It’s higher-performance, but we aggregate lower-performing components to a high performance level. If they fix hot-swapping and other issues it could be interesting.”
3D XPoint memory and flash in servers
We asked him if flash should be closer to the server? “It’s already been tried – and not adopted.”
”I think 3D XPoint in servers sitting in front of flash will be more popular than flash in servers sitting in front of disk with a greater latency gap ... Flash in front of flash doesn’t make a whole lot of sense ... Flash tiering doesn’t make a great deal of sense [either].”
“The XPoint memory could be a big persistent cache in front of external SolidFire flash.”
What does he think about the NVMe fabric idea?
“It’s for cases where latency is extremely critical, like some Wall Street financial trading, but that is a tiny percentage of the market. It’s very sexy but it’s not going to move the needle for flash.” Wright says SolidFire has more important work to do replacing disk.
He says SolidFire sees the latency and activity on its deployed arrays and there’s no gap between what array users need and what the arrays deliver. NVMe fabrics are expensive and complex and not generally relevant for SolidFire’s customers.
Intel’s Silicon Photonics ideas look more interesting, as it will be a standard and will offer 100Gbit/s and faster networking between server components like CPUs and memory.
In contrast, the NVMe fabric idea is obscure and proprietary. “How many NVMeF switches can you go out and buy?” he asks rhetorically. The NVMe fabric concept is taking the PCIe bus to places it wasn’t designed for.
NVMeF is also not as flexible as Ethernet.
Hyper-converged infrastructure appliances (HCIAs)
Wright said he thinks the hyper-converged idea can provide some benefits. Bringing compute to the storage layer is not so good for block-level access, but he sees value in doing it with files.
SolidFire customers can buy the Elements software on its own if they want a software-only storage product, and then bring their own servers, flash, and networking to it.
He wouldn’t say if SolidFire would go deeper into hyper-convergence: “We are in a target-rich environment [and] don’t want to get distracted chasing some shiny object in some other part of the market.”
Fresh funding and IPO
Asked about the possibility of a fresh funding round, Wright said; “An IPO is a funding round [and] we will have an IPO.” He’s waiting to see how other IPOs progress. We think of Pure Storage here which has already filed its S1 IPO papers.
SolidFire is rooting for good news with other storage IPOs, as that could encourage it to pull the trigger on its own IPO process.
If we assume that a $100 million revenue run rate is a qualifying hurdle for a SolidFire IPO, then we’re estimating that this could be reached by the end of 2016, with an IPO taking place, given favourable conditions, in 2017.
Software roadmap and summing up
There’s a 6–9 month software release cadence and SolidFire’s Oxygen release of its Elements SW came out in June. That would indicate a January–April 2016 timeframe for the next release. We can reasonably expect VVOLS support to be included, with Wright saying there will be other functionality in that release, but he’s not ready to talk about it.
There’s a SolidFire analyst day in February 2016, so we’ll probably hear a lot more then.
Wright said SolidFire’s business was split 50:50 between cloud service providers and enterprises. Enterprises are being added faster than CSPs, although CSPs tend to spend more in aggregate.
All this leaves us thinking SolidFire is a disciplined and, it sounds funny but seems right, conservative company, that is sticking firmly to its product mission and not having off after any new flashy thing, like NVMe drives and fabrics. It’s an all-flash company but not flashy, solid in fact; there’s a thought. ®