What are you looking for ?

Exclusive Interview With Lightbits Labs CEO Eran Kirzner

Accelerating storage with disaggregated architecture thanks to NVMe/TCP

Eran Kirzner is CEO and co-founder of Lightbits Labs, Ltd. launched in January 2016. Previously he spent more than 5 years at PMC-Sierra driving the software for enterprise storage, mobile and broadband activities following the acquisition of Wintegra where he worked almost 11 years leading some engineering and R&D development teams. Before that, he was at Freescale covering VLSI and architecture.

StorageNewsletter.com: How Covid-19 impacts you, your team and your business?
COVID-19 is affecting all businesses across the globe as companies reassess their immediate needs and budgets. At the same time, we’re also seeing companies realize they need to ramp up their digital transformation. Composable disaggregated storage solutions such as LightOS, our SDS solution, reduce TCO and improve flexibility, which is an integral part of digital transformation efforts as they help save money and time. We believe that challenges are an opportunity for growth and that together as an industry we shall overcome and come out of this crisis faster and stronger.

Share a bit the story behind Lightbits Labs, background of founders, the trigger of this adventure…?
Back at PMC-Sierra, I was running NVMe storage solutions and working closely with hyperscalers. Around that same time Avigdor Willenz (which I knew from Wintegra days) launched several companies in the infrastructure of data centers, compute, network, and storage space including Annapurna Labs in 2010, which developed the Nitro architecture. In 2015, Annapurna was acquired by AWS and Avigdor reached out to me about his next adventure. With Lightbits, our vision was to bring the flexibility, elasticity, and cost advantages of the hyperscalers to all data centers and that’s how our software, LightOS was born.

You promote a disaggregated model, is it the next storage wave especially at scale?
Absolutely. At Lightbits, we are strong believers in “going hyperscale or going home”. We think that any organization that is looking to thrive in today’s digital economy must look to the technology playbook of big cloud providers. As they have moved to storage disaggregation, the time has come for everyone else to follow, because the economics of disaggregated storage just make sense. 2020 is when we start to see the storage disaggregation ecosystem ramping up massively, including operating systems, NICs, SmartNICs and QLC SSDs while NVMe/TCP as the protocol to tie them all together. And that’s where LightOS comes in – enabling the next wave of SDS disaggregation.

You played a key role in NVMe over TCP development. Any details on that and why having NVMe now transported over TCP is so important?
As we looked at the state of NVMe-oF in early 2016, it was clear that the promise was huge but the current protocols for achieving it were leaving something short. They were either too old and slow (iSCSI) or too cumbersome to use (NVMe-over-RoCE). We pioneered NVMe/TCP so that anyone could deploy storage disaggregation at scale, while delivering performance that from an application perspective is indistinguishable from direct-attached SSDs. Virtually all data centers utilize TCP/IP today and people can take advantage of it now with NVMe/TCP. Extending NVMe across the data center over vanilla TCP/IP delivers the parallelism and low latency of direct attached NVMe without requiring any changes to existing network infrastructure or to application servers.

NVMe is a big change for the storage industry and invites vendors to reconsider storage infrastructure, the idea to transport NVMe over a network is pretty recent. It reminds me FC/SAN 25 years ago where users were able to share a storage array across servers over a FC networks. What are the differences beyond SCSI limitations?
You don’t have to compare NVMe to something as ancient and niche as FC, with apologies to all of the FC users and vendors. A more apt comparison we think is iSCSI. Just as iSCSI transported SCSI over TCP/IP, NVMe/TCP transports NVMe protocol over TCP/IP and can be viewed as a direct replacement – though one that can achieve up to 10X the IO/s at less than 25% the average latency of iSCSI and far better tail latency on the same hardware and networks. So the conceptual change is really not that big when compared to iSCSI. Just as iSCSI became commonplace because it enabled storage over the standard data center network (unlike FC), NVMe/TCP is standard, needs no special NIC hardware, and will be adopted readily as the data center storage fabric.

LightOS as a pure SDS transforming commodity server into a high-end storage array, give us the pitch about this.
LightOS is a SDS solution bringing hyperscale efficiency and flexibility to all. Any commodity storage server with NVMe SSDs can be turned into a storage powerhouse. It delivers composable, high-performance, scale-out and redundant NVMe/TCP storage that performs like local flash. At the heart of LightOS is something we call a Global FTL (GFTL). Unlike all other storage disaggregation solutions, it delivers storage features more commonly associated with high-end SANs: logical volumes, thin provisioning, compression, and data redundancy. It also extends the usable life of TLC and QLC SSDs, enabling the use of low cost flash without losing performance or latency. All of this is on standard servers and Ethernet networking at a fraction of the cost of traditional, proprietary arrays.

What about the clustered mode?
At cloud scale, everything fails. Any cloud-native storage solution has to be built from the ground up to tolerate – one might even say embrace – failure. So late last year, we introduced the industry’s first NVMe/TCP scale-out clustered storage solution, in LightOS v2.0. It protects against data loss and avoids service interruptions at scale. In the presence of SSD, servers, storage, or network failures, LightOS ensures applications simply continue working. Server failover and data re-balancing within the cluster is handled automatically, keeping data fully consistent and available no matter what fails. One unique feature of LightOS clustering is that a LightOS cluster can be spread across entire data centers. It’s not a rack-level solution or some giant appliance – a cluster is virtual and composed of servers located anywhere in the data center.

You also develop an acceleration card, LightField, what is the role of the card and for which use cases?
LightField is a PCIe hardware acceleration add-in-card (AIC) that enhances the performance, efficiency and utilization of storage resources. It integrates seamlessly with LightOS and provides a highly efficient hardware-based inline data-path design that improves system throughput, SSD utilization and endurance. Customers can get 100% of our data services and features in software-only with LightOS running on the storage server CPU cores, or they can use LightField to offload the CPU and thereby move to a lower-cost lower-power CPU option. For organizations looking to reduce TCO at scale, LightField is a great option.

You also sell an AFA named SuperSSD, why did you start this? Is it a PoC, a wish to show credibility but does it prevent partnership? Do you plan to continue in that direction?
We’re really excited about our SuperSSD appliance with LightOS inside, because it provides a full solution for those companies that want a complete end-to-end disaggregated NVMe/TCP-based solution. We saw that many customers and prospects wanted a scalable, Ethernet-attached SSD solution that specifically addresses the data storage speed, capacity and reliability needs of applications like AI and ML training systems. SuperSSD delivers that. In fact, a single SuperSSD based on the Dell PowerEdge R740xd 2U server can deliver up to 5 million IO/s with latencies under 150μs. That is, Lightbits SuperSSD outperforms the most expensive AFAs at a fraction of the cost.

How do you sell your product I mean essentially LightOS, the core of your business. In other words, what is your business model?
Just as LightOS is designed to fit the customer’s needs, our business model also accommodates the customer’s preferred purchase method. We offer our product as software-only, as software combined with a hardware acceleration card, or as a complete appliance via our SuperSSD offering. These offerings can be purchased directly from Lightbits, through solution integrators in our reseller network, or from strategic partners such as Dell Technologies.

Any OEMs and specific alliances?
LightOS is vendor agnostic and can be deployed on any server the user chooses to run on. In addition, Lightbits has an OEM partnership with Dell Technologies where we are currently promoting LightOS on Dell PowerEdge. We also work with International Computer Concepts, Inc. (ICC) as an OEM as our LightOS software runs on their STRATA Storage Appliance and expect to add more major OEMs across the globe by the end of this year.

Who are your competitors as NVMe and NVMe-oF is now a real red ocean?
The NVMe-oF space has really been heating up with several start-ups and even legacy AFA vendors adding the NVMe/RoCE protocol to their products. Lightbits jumped ahead of everyone by pioneering and promoting NVMe/TCP, avoiding the need for special RDMA NICs and the associated network switch/infrastructure changes. We’re happy to see that NVMe/TCP is now beginning to be offered by some of those same competitors and we see even more on the horizon. Since we are a direct replacement for iSCSI, vendors providing scalable iSCSI solutions can also be viewed as competitors. This competitive landscape is a great validation of the market need and we welcome the competition.

How do you explain that Israel is so strong in this topic of flash, NVMe and related protocols transport software?
It’s funny you should mention that. Did you know that Tel-Aviv contains more start-ups per capita than anywhere else in the world, even beating Silicon Valley in 2019? I think our country is unique in that it’s a “melting pot”, encouraging entrepreneurs to explore innovative technologies and constantly push forward the state of the technology. It’s not a surprise to me that all tier-1 players such as AWS, Google, Microsoft, and Facebook, have branches in Tel-Aviv; when the Israeli entrepreneurial spirit combines with the innovation discipline of the hyperscalers, it creates magic. I think innovations in flash, NVMe, and protocols such as NVMe/TCP are a direct result of the Israeli mentality and circumstances.

What is your cloud strategy?
LightOS is a natural fit for anyone looking to build an exabyte-like environment in their private clouds. We’re fully aware that many customers are interested in both hybrid cloud and application portability between public and private cloud and would like to run LightOS in their public clouds of choice. This is something we are working on at the moment. Watch this space for updates.

How do you work with containers?
LightOS really delivers on the promise of containerized cloud-native infrastructure. As you know, Kubernetes is the leading container orchestration platform in use today. It was initially designed for DAS and gained support for disaggregated storage along the way. Today containerized applications can enjoy all of the benefits of disaggregated storage using Kubernetes support for CSI (Container Storage Interface), which LightOS supports. With the LightOS CSI plugin, you can spin up persistent storage volumes for your containerized workload just as easily as you spin up another container. And if you happen to use some other orchestration platform for your VM or containerized workloads, such as OpenStack or OpenShift, LightOS supports them too.

You raised so far $50 million in one round in 2019 with Micron, Dell Technology Capital and Cisco among others, how did you bootstrap your company, any money from early investors? I see Avigdor Willenz as the chairman of Lightbits Labs.
We’ve raised $55 million over 2 rounds. Our investors include Dell Technologies Capital, Cisco Investments, and Micron, as well as top investors and VCs, such as Avigdor Willenz, Lip-Bu Tan, Marius Nacht, SquarePeg Capital, and WRVI.

Employees, offices, revenue?
We’ve brought in an experienced C-level team, all of whom have prior experience with leading storage, networking and hyperscaler organizations, including Facebook, Google, DellEMC, MLNX, PMCS, and others. We are just shy of 100 employees with offices in Silicon Valley, New York, Shanghai, and Israel, and we’re continuing to grow.

Read also:
Lightbits LightOS Delivers NVMe/TCP Clustered Storage Solution
Protects vs. data loss and avoids service interruptions at scale, and maximizes operational efficiency, ensuring applications continue working in presence of failures, and failover is handled automatically, keeping data consistent and available
December 20, 2019 | Press Release
VMworld: Lightbits Labs Demonstrated How LightOS Solution Working With VMware vSAN
NVMe/TCP brings disaggregation to VMware vSAN HCI, scales storage and compute independently
September 3, 2019 | Press Release