Interview: Whamcloud's Chroma Puts a Shine on Lustre Storage

Print Friendly, PDF & Email

Today Whamcloud announced its new Chroma management software for the Lustre file system. I saw an impressive demo of the Chroma beta back at ISC11 this past June, but now it’s ready for prime time. To find out more, I caught up with Robert Read, head of the New Products Group at Whamcloud.

insideHPC: What are you announcing today? What is Chroma?

Robert Read: Chroma is a purpose built system for virtualizing storage using the Lustre filesystem, including a new management console for easy configuration, maintenance and monitoring of clusters. It is designed to utilize enterprise-class storage for top performance with minimal interruptions. Chroma will be available through Whamcloud partners as a Lustre appliance.

insideHPC: Is this a first step towards bringing Lustre to the Enterprise?

Robert Read: Actually, the first step to bringing Lustre to the enterprise was to significantly raise the bar for Lustre stability. Lustre’s speed and scalability is well-known. But enterprises demand stability and dependable release schedules. Whamcloud and the Lustre community have been focused on stability and dependable releases over the last year and Lustre 2.1 came out this fall continuing this emphasis. Chroma is the next step. We’re using the core Lustre file system as the key component of a complete scalable storage solution.

insideHPC: Lustre has a reputation for being difficult to set up and maintain. Is Chroma designed to address these issues?

Robert Read: Yes. Chroma is deeply integrated with Lustre, and so it’s more than just “making Lustre easy.” Though that’s certainly a selling point. Chroma vastly simplifies installation, configuration, maintenance, monitoring, and fault diagnosis. It provides great insight into your storage system that otherwise would require exceptional Lustre skills and a serious do-it-yourself investment.

insideHPC: Will Lustre experts use Chroma instead of their trusty scripting, do you think, or will it get in the way?

Robert Read: I believe Lustre experts will use Chroma as they will love the amount of information it gives them about the Lustre storage systems. However, for this initial version the target customers generally have less complicated deployments and do not need heavy Lustre experts, but still want many of the advantages of utilizing Lustre. So one of our goals for this initial version of Chroma is to make Lustre manageable by non-experts.

insideHPC: What do you say to people who say they “don’t need a GUI” for Lustre?

Robert Read: I agree with them. You don’t need a GUI for creating or managing Lustre filesystem, though the live graphs are handy for real-time monitoring. What you do need, or at least what would be extremely helpful, is a centralized system for managing the storage and file system configuration. For instance, it can be troublesome to manage shared storage when you mange servers individually, but a central management system can detect all the shared devices and present them in a much more intuitive way. Chroma is about managing storage virtualization, with emphasis on Lustre, of course. The GUI is one interface for that, and naturally we’ll have a CLI as well.

insideHPC: How does Chroma compare to other management consoles out there?

Robert Read: I’m not an expert on all other management consoles, but a common approach I’ve seen is packaging a variety of existing solutions under one umbrella. The end result with this approach often leads to having silos of information, where the file system configuration is separate from the storage configuration, and performance stats are stored in a yet another place.

Chroma is a completely new system built from the ground up, and all aspects are tightly integrated. For instance, when we receive a notification from the storage subsystem that a particular disk is about to fail, then we know exactly which Lustre services are impacted. We can bring together information from multiple sources, such as storage events, Lustre metrics, system logs, etc., so the administrator has a unified view of what is going on in their storage system.

insideHPC: How is Chroma different from open source Lustre?

Robert Read: A large component – and a lot of the value of Chroma as a storage virtualization solution – comes from Lustre itself. What makes Chroma different from the community release of Lustre is the integrated management system that ties Lustre together with the underlying hardware. Chroma encapsulates provisioning, management and monitoring of Lustre filesystems in a single API and user interface. It is highly customizable and allows for plug-ins for custom actions and storage backends and an API for integration with third-party management tools.

insideHPC: Do you have a feature roadmap? What’s around the corner for Chroma?

Robert Read: We will definitely be extending Chroma to handle extreme scale deployments. We’re also very keen to solve large scale data management issues, and we’re exploring ways to provide a truly scalable single namespace solution to cloud applications. If Chroma would be helpful in your clustered environment, now is a great time to contact us and have a discussion. We would love to hear your issues and include your input.