Panasas has a new CEO to lead the company into the AI boom.
Founded in 1999 and headquartered in San Jose, Calif., Panasas sells parallel file systems -- storage systems that distribute file data concurrently across servers to increase performance and scalability, mainly used for high-performance computing (HPC) environments. New CEO Ken Claffey, formerly the general manager of enterprise business at Seagate, is aiming to guide Panasas into cloud-centric, data-centric terrain.
In this Q&A, Claffey, who founded Seagate's HPC business that was sold to Cray and operates as ClusterStor today, talks about the change in leadership for Panasas, explains how the company is positioning itself to compete in today's storage market and makes the case for the role of parallel file systems in generative AI.
Editor's note: This Q&A has been edited for clarity and conciseness.
You are Panasas' third CEO change in three years. Why is the board bringing you in now and moving current CEO Tom Shea back to the COO role?
Ken Claffey: Tom [Shea] stepped in when the previous CEO, Faye Pairman, left. ... He did some really great work in terms of steadying the ship during the transition. The board felt it was the right time to bring in a different CEO that might be better positioned to enable the company to take advantage of the opportunity in front of us, going into a growth mode.
Tom is in the office next door to me and is staying on as COO. I highly value the work he has done and the expertise he brings to the table.
When you say 'growth mode,' what do you mean and how are you ramping up?
Claffey: If you look at where the company is positioned right now, there is more demand for the technology and capabilities than we've ever had before, largely linked to the growth of HPC and the growth of AI. [Panasas] has been growing incrementally over the last few years as it has made the move to get off proprietary hardware. We are a software company. All of our engineers are software engineers.
[Our new outlook] is about making our technology available to customers in different ways. For consumption, if you want an on-prem appliance model, we can support that. If you want to consume and get value out of our technology through different mechanisms, be it public or private cloud, or linking it to as-a-service models, that is how you will see Panasas evolve in the coming years.
Panasas doesn't have much of a public cloud presence, correct?
Claffey: It does not. This is linked to how [Panasas was previously set up]. When you have a software stack that is tightly tied to a very bespoke hardware environment, that precluded you from moving to the cloud and offering your customers that capability. As we go forward, that is something that we will be adding to our portfolio.
Panasas supplies storage for HPC used in research. Earlier this year, the company expanded to edge locations. Where are you going next?
Ken ClaffeyCEO, Panasas
Claffey: Unlike a lot of traditional HPC companies, the vast majority of Panasas customers are blue-chip enterprise accounts. ... As more enterprise companies look to adopt more HPC-like compute clusters as they start to deploy their AI models, that's a growth opportunity for us. Taking the capabilities of a parallel file system, but packaged in a way that is easy to deploy, mange and maintain as more customers adopt GPU-driven technology.
I think you'll see a trend of HPC technologies being deployed to support generative AI applications and workloads. If customers are going to do that, at any scale, they will need a parallel file system such as Panasas that brings enterprise-class performance and scaling.
What does a parallel file system bring to the generative AI table for storage?
Claffey: If you are doing generative AI, you have very complicated models that will need a lot of compute power, largely GPUs. Looking at a parallel file system versus a traditional NAS system for a big compute cluster, the parallel file system can provide performance for 20, 50, 100 different servers. The one NAS box will become a bottleneck. Panasas' parallel file system allows you to concatenate the performance of multiple storage boxes, essentially marrying a storage cluster to the compute cluster.
This was the problem of building a supercomputer that parallel file systems solved. Unlike other storage technologies that were never designed from the ground up to scale out, parallel file systems are, by definition, parallel. They are designed to scale.
More broadly speaking, does storage need to support use cases such as generative AI?
Claffey: The way the storage world works, as applications change or have different I/O workload requirements, the underlying storage layer needs to react to that. Storage will continue to evolve as it always has.
Adam Armstrong is a TechTarget Editorial news writer covering file and block storage hardware and private clouds. He previously worked at StorageReview.com.