Garth Gibson, is one of the authors of the original RAID paper (pdf), CMU professor, founder of the Parallel Data Lab, founder and head of the Petascale Data Storage Institute and founder and CTO of Panasas, a maker of parallel clustered NAS systems. I caught up with him at the Seattle Scalability Conference in June and taped about 40 minutes of conversation.
After much procrastination I got the videos edited and up on YouTube a month ago. In 16.5 minutes Garth covers large scale file systems, massively parallel supercomputer failure modes and backup strategy. Panasas provides the back-end storage file server for 6 of LANL’s supercomputers, including the world’s fastest: a 1 petaflop sustained machine named Roadrunner.
Here is part I:
And here is part II:
The StorageMojo take
The problems that LANL is having today will be much more common in 10 years. Not that we’ll all be running huge informatics apps or 6 month simulations of nuclear weapon decay, but if they can figure out the software, many of us will have 64 core – or more – desktop systems equal to a respectable commercial HPC installation today.
Maybe we’ll be running personal informatics jobs, looking for nascent trends in, oh, I don’t know, fantasy sports or parallel NFS. Or hosting virtual 3D game worlds that keep evolving even when we’re away. Somebody will think of something cool to eat those cycles and terabytes.
Courteous comments welcome, of course. I did some work for Panasas last year. I also like their leadership on parallel NFS and object-based storage.