WekaIO: Accelerating Business Outcomes with a Limitless Data Platform

They get the best benefits of both platforms and the applications have immediate access to all the company data in one file namespace. When it comes to compute [GPU] acceleration, after being in effect for decades, Moore’s Law [the periodic doubling of processor performance] is dramatically slowing. Accelerating AI/MLoutcomes hinges on the modernization of processors to increase performance. We are seeing more enterprise organizations making the switch from CPU to GPU computing technologies to accelerate their data analytics workflows, especially for AI/ML use cases. What most organizations fail to realize is that when they employ GPU-enabled servers that their I/O stack needs to change too. Most ML initiatives fail because the storage platform has not been considered when GPUs are employed, and these expensive data-hungry servers just sit idle and wait for data to get to it. GPUs are a great investment for organizations that want to accelerate ML training models and get products to market faster, but unless the storage platform is considered as well then the return on investment isn’t maximized. Because it’s flash-optimized, cloud-native, and delivers the best I/O throughput to GPUs, the Weka Limitless Data Platform is the best solution for ML workflows.

What is your company’s role in data center digital transformation, and what is the value that your product brings to the table?

Weka is a true scale-out NAS, unbounded from the capacity limitations of legacy NAS. Our storage solution delivers utmost scalability and is ‘limitless’ in the real sense. Customers with ML workflows love our storage because they can access their massive data sets in one file system namespace—managing hundreds of billions of files in a single directory. In that sense, you could draw similarities to efficiencies of object storage, but with the added benefits of high-performance and ease-of-use.In contrast to first-generation flash-based solutions that introduced half of the storage as flash connected through PCI to increase performance, Weka demonstrates how a shared storage architecture can deliverbetter speed than block (SAN) storage. And while single application workloads work well with traditional second-generation flash solutions organizationsmust makechoices around tuning [capacity, performance] parameters and when new applications are introduced the system ceases to work effectively. Weka eliminates the need for tuning parameters and instantly adapt to any new workload—without any knobs to tune or expertise.

Weka eliminates the data silos persistent in legacy storage architectures—if the application can’t access the data then it offers no value to the organization. Wherever your data lives, Weka manages that data in a global namespace ensuring the applications have access to all the data for better business outcomes. In addition, you can manage your data anyway you want because Weka is deployable on-premises, in the Cloud, or a hybrid—it’s your data, managed your way with no vendor lock-in.

Weka is software-only and available on any server vendor of your choosing; it’s acquired as a total solution through partners in the Weka Innovation Network™ (WIN). Again, this eliminates vendor lock-in while delivering all the enterprise features, including synchronized data security over Windows and Linux. In essence, we are the most flexible storage solution on the market today that delivers enterprise simplicity, speed, and scale that complex ML projects require.

While GPUs are an important aspect of such projects, we focus on the most important component—the data. We ensure that our clients can store the right amount of data and drive the right amount of data into the GPU. For example, an electric vehicle is creating a difference not because of the electric engine that has been in use since the eighties but for the advances in battery technology. Similarly, the AI and machine learning data is more than the machine; if you just had the engine, there is no drive and no learning. Weka is like the battery that enables the vehicle to view data as an effective component of learning.

Please share a case study or two to help understand the benefits that Weka brings to its clients.

Weka was approached by an organization in the autonomous driving vehicle market. The client was using a second-generation flash product from a well-known company and it wasn’t meeting their expectations with regards to speed and scale—it was taking too long to complete their training epochs and delaying time-to-market. With Weka, the clientshortened their single training cycles from two weeks to a few hours. This allowed them to achieve their goals and accelerate their product development within a week; such impressive results allowed the client to maintain a competitive edge in the market. And without have to retune the system, the client was running run several different applications on the same storage platform efficiently.

Another client with a similar use case was looking at us to help improve their GPU utilization and reduce their model training time. The client had decent compute capacity, running on a steady state. However, once every week, the company aimed to run a longer and bigger training model over more resources. Weka was able to save the client money with an on-premises implementation, while also producing snapshot copies and replicating the data to AWS S3. This workflow allowed the client to run large-scale model training projects by just migrating their workloads to AWS. The workflow with Weka provided scale, speed, and elasticity that did not introduce latency or additional time to move the data. As a result, the client achieved better business outcomes at a much faster speed and better cost economics. This client was also able to grow top line revenue by accepting more business; due to the accelerated speed with which training cycles were completed they were able to accept more jobs.

What are some of the significant differentiating factors that keep Weka ahead of the competition?

We have a convergence of an extremely talented and experienced engineering team who have architected a highly innovative solution and the fact that we are software-only. We are unique solution that offers the most flexible deployment options—choose your server and choose your platform because Weka works anywhere. Users can expand capacity without any limitations while the system is running, with no performance degradation, and there is no proprietary hardware.

What does the future look like for Weka?

We envision both the use of Kubernetes for containers and the Cloud to increase in user adoption—and so for us from a technology perspective, we will invest in R&D resources to build complementary features for both use cases. We are leveraging Kubernetes to unify orchestration and make the run time environment transparent to applications. And we will diversify our Cloud solutions to support both GCP and Azure users. And we announced our partnership with Hitachi Vantara last year, the product will GA this year, which supports our global expansion strategy.