Storage industry veteran Mike Gustafson became CEO of PCIe flash card startup Virident Systems in September 2012 after spending six years as CEO of NAS vendor BlueArc, which he sold to Hitachi Data Systems (HDS) in 2011. Gustafson also spent a year with HDS after the acquisition, and was a senior vice president at Fibre Channel switch vendor McData for six years before going to BlueArc.
Gustafson's first six months at Virident have been active for the startup. In January Virident scored an OEM deal with Seagate to sell its FlashMax II cards, and rolled out its FlashMax Connect caching and management software suite in February. And while neither EMC nor Virident have disclosed a partnership, it's an open secret in the storage world that EMC is selling FlashMax II cards as part of its XtremSF family of PCIe flash.
We recently spoke with Gustafson and Virident CTO Vijay Karamcheti about Virident's technology strategy and "software-defined flash."
After bringing out your latest card, the FlashMax II, last August, Virident has concentrated on software for caching and pooling storage across servers. You also talk about 'software-defined flash.' Do you see Virident as more than a PCIe flash card vendor?
Mike Gustafson: We expected from the beginning to be more than a card in a server. We always expected to have it in shared storage, and bring the benefits of shared storage onto the server side.
What role do you ultimately see flash playing in enterprise storage?
Gustafson: Flash has been a well-known technology for low-latency and high-bandwidth applications, but we're starting to see it become more pervasive. Core enterprises are starting to figure out what they want to do with this architecture, one, two, three years down the road. They're looking at whether the software is deliverable today and over time, to where they can truly use this as a shared performance tier.
We think there will be a performance tier dominated by solid-state flash, and a capacity tier that remains spinning disk and either private or public clouds. To get there, there's a ton of work that has to be done by the industry.
We have vHA, vCache and vShare software [as part of the FlashMax Connect] suite. You have to look at how to add high-availability attributes and replication for mission-critical applications, and how you begin to add caching, not just isolated in the card but through all the storage tiers from the server through the array. With vShare, we start to get attributes of shared storage.
Vijay Karamcheti: So, there's a set of capabilities that we have now, but we need to make it integrated in a way so applications can leverage those capabilities inside the server and closer to the applications.
Will Virident's focus be on software from here on?
Gustafson: Hardware is still important, but only to the extent where the software makes it intelligent.
You can't just take flash, stick it on a card and call it a product. With six years of [development] experience, we have a lot of scar tissue and understanding of storage-class memory. There is a deep set of IP and scar tissue that's helped us optimize the software for what we call the 'flash translation layer' and 'flash management layer.'
The majority of our developers and intellect has been focused on the software side since the company's inception. There's a lot of magic involved when you can pack 2.2 terabytes on a half-height card, and that can't be taken for granted. But software has been a focus for us from the beginning, driving unconditional performance, a tight band of predictable consistent performance from Day 1. All of those aspects are software-centric.
There are three phases to our software strategy. First, there is the foundational aspect of how you manage flash. Then, how do you take that to the next level so it's tightly coupled through open APIs, and then get to deeper interoperability and integration with the application layer. We're well down the road, but there's a lot more work to do.
You expect flash software to be open? Do your competitors agree?
Gustafson: On principle it must be open. Customers expect it to be open. We architect it to be open. We do, however, have a huge advantage with our two years in flash. We have the flash translation layer to manage and tightly couple hardware -- including the flash -- and software. We optimize based on what we know about the uniqueness of flash and what we can do to make it sing, but we also understand it has to remain open.
From a business standpoint we're going to lead with our own capabilities, but we're absolutely committed to be open. Large enterprises expect that from us, and I don't think all of our competitors share that vision.
Gustafson: There are three pillars of the business. One is the products and technology. The second is partnerships and [the third is] the business model.
We don't fear Fusion-io. We have a great deal of respect for them. Our technology is strong. What we lack is a level of revenue and share growth commensurate with having that technology. We're excited about the Seagate announcement and the addition of other partners. We expect to grow two times the growth of the market this year. We don't disclose our revenue, but we beat our plan considerably in the fourth quarter and are beating it in the first quarter.
Is the price of flash low enough for enterprise adoption to really take off yet?
Karamcheti: Price points are getting to where customers are thinking about what we call pervasive flash. Look at some higher-end applications like Oracle; it has been cost effective for the last six or nine months. To get that level of performance, you need a lot more drives to support the I/O so the overall cost of the configuration tends to be higher with spinning media than flash. VDI [virtual desktop infrastructure] is another example of this.
Using lower-cost [multi-level cell] flash with software to make it reliable with high endurance has made MLC the primary technology across the portfolio today.
Does single-level cell [SLC] flash have a future?
Karamcheti: We don't see SLC going away, but it will probably get relegated to a small subset of applications that have huge write requirements. Of the applications we encounter, 95% to 99% do not write as much to require SLC. The demand is satisfied with MLC combined with our software technology.
Will Virident sell TLC [triple level cell] flash any time soon?
Karamcheti: We're definitely looking at it. At this point, in terms of the applications we are going after, we don't see a dominant use for TLC yet.