Exclusive Interview With Gene Banman, CEO, DriveScale

Gene Banman is:

• CEO of DriveScale, Inc.
• Board member at Curriki
• Board member at Zero Motorcycles Inc.

He was:

• CEO of Clear Power Systems (acquired by Questor Technologies, Inc.)
• CEO of Zero Motorcycles Inc.
• CEO of NetContinuum (acquired by Barracuda Networks)
• President at Mirapoint Inc.
• VP and GM desktop business unit at Sun Microsystems Inc.
• President at Nihon Sun Microsystems, KK

StorageNewsletter.com: DriveScale is a company pretty new for many readers. Could you share with us the genesis of the project with of background of founders as many of them spent some time at Sun Microsystems?
Gene Banman: Satya Nishtala and Tom Lyon, the founders of DriveScale, worked on several ideas around distributed computing and realized that one of their ideas, disaggregated storage, had real potential to bring significant value to data centers running Hadoop/Spark and other distributed computing environments.

While at Cisco as chief designer for the UCS product group, Satya read the papers published by Google about their distributed computing architecture and the MapReduce programing environment they were running on. He saw that this type of distributed computing approach running on commodity servers had tremendous advantages in terms of cost, throughput and massive expandability and that this represented the future of computing.

He recruited Tom, a good friend and founder of Nuova Systems (the startup which was purchased by Cisco to become the UCS product group), to join him and evaluate several of his ideas on how to improve this distributed computing platform. Tom and Satya had also worked together at Sun Microsystems, where Tom has been an architect of the NFS and wrote much of the Sun networking stack and Satya had been the chief architect for Sun’s workstation product line.

You raised only $17.5 million in 2016 which is pretty small, have you a frugal approach with investments?
Small teams can be very productive, and we have kept the team small while we validate our market fit. We have some major opportunities closing this year at which point we will raise more funds and continue expanding.

DriveScale plays in a new infrastructure category, could you summarize the vision of the company?
We provide Software Composable Infrastructure for modern workloads. This means our value is in orchestration software for the distributed computing platforms we are targeting as described above. Since the hardware has become standardized and commoditized, the next generation of computer company value add will come in this orchestration capability: managing the composition and interconnection of standardized compute, networking and storage resources to most effectively handle these modern distributed computing workloads.

Is it a pure software product?
DriveScale’s Software Composable Infrastructure product suite connects storage resources to compute resources to ‘right size’ servers and clusters of servers in order to precisely match the requirements of distributed workloads. With the traditional approach of utilizing commodity servers with direct attached drives, the amount of storage per server in a cluster was fixed at purchase time and stayed the same through the life of the server. With our approach, storage is allocated as needed and can be changed whenever the workflow requirements change. In some situations, a bit of DriveScale hardware is required to connect commodity storage in JBODs to the Ethernet switch and then on to the servers. In other cases our solution is wholly software and can connect smart JBODS to the Ethernet directly.

What are the motivation of users adopting DriveScale, what do they wish to achieve?
There are several major advantages of disaggregating storage from compute in these distributed computing applications:

• The user can allocate just the needed amount of storage and I/O to the cluster, knowing that he can adjust as necessary at a later date. Without this adjustment capability, users typically over provision substantially to account for changing requirements.
• The user can manage the life cycle of compute and storage on different time frames, upgrading processors as needed without throwing away good disk drives.
• Multiple distributed applications (clusters) can share the same resources; e.g. a MapReduce cluster with 12 drives per node; a data lake with 50 drives per node; and a Spark cluster with 2 drives per node, can share the same infrastructure. With the ‘silos’ gone, processors can be moved around between these applications as workflows change yielding superior capital utilization.
• Server failure can be handled quickly and gracefully without rewriting all of the data which was controlled by that server. Its drives instead can be assigned to a backup server or redistributed across the cluster, recovering the data from that failed server in minutes instead of hours or days.
• All of this is accomplished with no change to the application stack or to the performance behavior of the application. Everything runs as if the drives were direct attached within each server, but with the flexibility of attaching the drives through the top of rack switch.

What is the minimum configuration?
These distributed compute applications typically utilize hundreds of servers. However, a minimum cluster could be put together with one JBOD and 6-10 servers for a small cluster.

How do your price your solution?
We charge a subscription fee for each server node and each drive, per year. If our hardware is needed in the solution, we sell it for a one time charge. We also offer extended service offerings. Business hour support is included in the subscription fees.

What is the addressable market?
There are approximately 1,000,000 Hadoop/Spark nodes in the market. There is a similar number of other distributed applications nodes running No-SQL or server storage-based software. All of these nodes and clusters are a target for our technology.

What about Europe and APAC for such technologies?
In our first year of shipments, we already have strong prospects for our product in China and Japan. We have completed one POC in China and have a commitment for a POC in Japan starting early next year. We know that distributed applications have wide penetration in Europe and we will be entering the European market in 2018.

How big irons react to such approach? I mean Dell, HPE, IBM and other commodity servers companies?
They have been quite receptive to working with us. We are in partnership programs with HPE, Dell EMC and Cisco. We are in the Peripherals and Storage Dell EMC price list, and our Dell customers can order our products from their Dell EMC sales rep.

By making clusters built from their products composable, we are helping their customers provide cloud-like flexibility to their users which in turn helps them keep these applications in-house.

What are the preferred uses cases for composable infrastructure? It seems that it’s a good fit for frequent changing configurations so I imagine telcos, service providers, etc?
Really anyone running large distributed computing clusters is a good prospect for us. Two public customers are AppNexus and ClearSense. AppNexus is the largest independent AdTech company (Google’s DoubleClick is the largest) running thousands of nodes in multiple data centers, serving billions of ads every day, and needed a solution that would let the company scale its data center quickly and efficiently. Our SCI technology has helped AppNexus significantly decrease operations workloads and reduce resources trapped in siloed clusters to meet its scalability and flexibility needs.

ClearSense is a real time medical analytics company providing predictive analytics on hospital patients to improve their care. It is a fast-growing start-up and is running their analytics clusters entirely on DriveScale’s SCI platform. Composable Infrastructure enabled Clearsense to move from AWS public cloud to an on-premises private cloud environment, saving significant costs and maintaining the cloud-like agility they were accustomed to.

In term of storage, what storage flavor do you support? Any plan for NVMe and NVMef?
Currently we support SAS HDDs. Stay tuned.

What about the competition?
Our primary competition is to just use standard commodity servers with direct attached drives. Other disaggregated storage solutions are way too slow and expensive for use in these large distributed computing applications.

What is your business model? Do you plan any OEMs?
We have several OEM prospects and are active in licensing DriveScale’s SCI technology.

How do you finish 2017?
This was our first year of shipping, and in 2018 we’re anticipating 10x growth.

What do you plan to add in the product in 2018? and for the company, a new round to accelerate the development?
We have some interesting new products and new platforms supported in 2018. Stay tuned.

Read also:
Start-Up’s Profile: DriveScale
In data center orchestration software for big data
by Jean Jacques Maleval | 2017.03.15 | News