Any baseball announcement from EMC typically gets my attention. The MLBAM press release was no exception.
MLB Advanced Media (MLBAM), the interactive media and Internet company of Major League Baseball, was looking to improve its video streaming and e-commerce capability to baseball fans the world over.
MLB.com is one of my favorite websites. MLBAM, which happens to be a limited partnership of the club owners, operates the site.
I was in Russia recently when I saw my son Tweet about the Dodgers/Padres brawl. Within 30 seconds I was watching the streaming video of the interviews after the game, courtesy of MLB.com. According to Wikipedia the site gets about 4 million hits a day. Fans like myself want two things: 1) Stats, and 2) Video. Many fans take this a bit farther and consume massive amounts of streaming video.
As the hunger for video, streaming content, and lightning-fast statistics grows to a global scale, the MLBAM IT infrastructure needed a refresh. The file systems were aging, the implementation was spread across multiple heterogeneous systems, and upgrading the complex, disparate systems meant downtime.
So they came up with a two-pronged solution:
- Let's store the video files on one of the fastest, largest, robust file systems in the world.
- Let's store the statistics on one of the most massive, reliable rack of raw storage we can find.
The announcement highlights Isilon as the file system choice and the VMAX 10K for the raw storage. I've written quite a few posts on VMAX; the MLBAM implementation of the VMAX 10K uses virtual provisioning to lower the overall storage needed for the initial roll-out. The VMAX not only serves up statistics, it stores the main database that allows MLBAM to generate revenue via e-commerce.
For this post, however, I'd like to focus on Isilon's metadata acceleration feature.
When I click on the Dodgers-Padres highlights, my request translates into a "file open" and subsequent read request. Most file systems, when they receive an "open request", perform a small number of disk operations to retrieve metadata about the file. This metadata acts as a bread crumb trail to the actual physical location of the video I want to watch.
Even a small number of metadata lookups to disk results in some serious delays at global scale. This is where metadata acceleration comes into play. Instead of storing the metadata on spinning disk drives, it is stored on solid state media. My colleague Jonathan Rozinsky put it best:
From an Isilon perspective, metadata acceleration is the use of SSD to store copies of file system metadata (mtime, ctime, atime, filename, file size, etc) and metacontent (inodes) in order to increase the speed with which the file system can respond to read requests. This is especially helpful in the case of small file, high i/o, random read intensive workflows such as virtualization, OLTP and large scale home directory/file repositories. However the benefits are not limited to such scenarios. We have seen performance benefits in almost every type of workflow when metadata acceleration has been applied, and our clients continue to find new ways to leverage this technology to improve the speed of their workflows and decrease their overall time to market.
I recommend you connect to MLB.com and peruse the site. Observe the speed of the video capability. Isilon metadata acceleration at work!