Return to Direct Attached Storage for DreamWorks
Is IBRIX truly a vendor-agnostic technology that can dramatically improve data access speeds?
Milan Shetti, vice president of marketing and business development for IBRIX, sounds as though he is ready to walk down the red carpet to pick up a technical Oscar award from the Academy…and he would like to thank all of the direct-attached arrays and internal server disk drives that helped make it possible.
In DreamWorks Animation SKG, he has found a customer that is tired of the capital and operating expense of its storage area network, especially as a platform for facilitating the needs of animators who are working with tight deadlines to produce feature animation films such as last year's Bee Movie and soon to be released Kung Fu Panda. Note: If you don't know these titles, ask a child.
According to Shetti, about a year ago DreamWorks began asking the question, "Why not just use servers with local storage -- direct attached disk and internal drives -- to get images rendered more quickly?" From a CAPEX perspective, such a strategy would result in a five- to ten-fold cost savings by eliminating the complexity introduced by a Fibre Channel fabric.
What such a solution would need, of course, is a way to manage access to a substantial amount of data stored on direct attached islands. Early Fibre Channel enthusiasts saw their version of a SAN as the solution: detach all arrays from the servers, pool them together behind a simple (but expensive) switch, create zones to describe which switch ports would provide access to which storage elements, then cobble the ports to host bus adapters on the servers. Several users, therefore, would be able to share access to the same storage, and storage itself could be managed horizontally as an infrastructure layer.
Of course, building such an infrastructure, as DreamWorks learned from the deployment of its own FC fabrics from Hewlett-Packard, requires considerable time and expense for wiring, programming, and administration. They weren't prepared to abandon the investment already made in a 500 TB fabric, but they went looking for a new approach to streamline their primary work, especially to handle intensive rendering of animation sequences. That's when they discovered IBRIX.
IBRIX technology is deceptively simple in concept: it creates a global namespace that virtualizes file systems and storage elements. This engine parallelizes access to stored data from the server point of view, enabling high-performance, concurrent access to files wherever they exist. IBRIX supports any storage platform, from internal disk to SAN, but doesn't depend on any particular configuration to work. In the case of DreamWorks Animation, it provided the means to access local and direct attached disk with greater alacrity and at lower cost than accessing the same capacity deployed in a costly and difficult to manage FC fabric.
Anyone close to the film industry understands that profit margins can be as tight as schedules. A single animated feature uses over 100TB of data, consisting mostly of small files of just a few KB. From a CAPEX cost perspective, DreamWorks wanted to buy that capacity at the lowest possible cost, preferring DAS and internal disk to other, more expensive storage configurations. HP accommodated them with ProLiant DL385 servers and HP StorageWorks MSA70 disk shelves -- and, to make it all accessible and efficient, IBRIX Fusion file servers (essentially a cluster storage head running the IBRIX engine software).
Such a configuration, although dramatically less expensive than a SAN, also needed to facilitate the high-speed access and compute requirements of DreamWorks' animators. Shetti explained what's involved: "The rendering process of the movie is a very compute-intensive application and is executed in multiple phases of the movie production. At DreamWorks, for just the lighting part of the rendering process alone -- there are currently over 17 million small files in a single rendering working data set."
Using SAN-based access, which was replaced by the HP-IBRIX solution, artists'' editing cycles and ultimately the creative process as a whole were slowed by the need to wait overnight for high-resolution renders. According to Shetti and Ed Leonard, CTO at DreamWorks Animation, the new configuration, leveraging local disk and global namespace services from IBRIX, provides significantly higher I/O performance "allowing artists to work up to five times faster. As a result, higher quality scenes with more accurate lighting detail can be produced faster and with more artistic control than ever before."
"This storage solution not only allows artists to see their high resolution renders within minutes but helps us launch an entire generation of new, interactive tools that we hope will raise the bar on animated film quality," said Leonard.
To IBRIX, this case study is another feather in the crown and another prestige customer on a growing list that already includes prominent Wall Street risk analysis firms, telecommunications companies, film and media companies, pharmaceutical houses, and oil and natural gas firms. Not all have abandoned FC fabrics as their preferred storage platform, but those that have chosen to adopt the strategy that DreamWorks is following in their mission-critical rendering process have gleaned the double win of lower infrastructure cost and faster access speeds.
It is also proof that IBRIX technology is vendor agnostic. While many of its past sales have been made through partnerships with EMC and Dell, DreamWorks Animation is their first partnership with HP. In the future, they hope to work with HP to enable the hardware company''s forthcoming SAS-SATA arrays with an internal age-based data migration scheme based on IBRIX name spaces.
Watch this space for more information as the IBRIX story develops. It will be worth a look.
Your comments are welcome. firstname.lastname@example.org.
Jon William Toigo is chairman of The Data Management Institute, the CEO of data management consulting and research firm Toigo Partners International, as well as a contributing editor to Enterprise Systems and its Storage Strategies columnist. Mr. Toigo is the author of 14 books, including Disaster Recovery Planning, 3rd Edition, and The Holy Grail of Network Storage Management, both from Prentice Hall.