“NOTE: The below post was originally written and published on GestaltIT.com as part of a sponsored content series with Pure Storage.”
I’ve done some research into POSIX storage (Portable Operating System Interface) which is really centered around the High-Performance computing world, and works most well in Unix environments. In concept, a POSIX environment specializes in the concurrent reads and writes into the storage environment. In HPC, the world must rely on a couple of key details, those are very fast transactions, and of course scalability. The limitations have always stemmed from the speed of disc, and the interconnect on that side, in terms of performance. Also, a key factor in the relevance of these file-systems is how the metadata is leveraged such that performance isn’t limited. Then, in terms of scalability: while there’s been much done, the growth in storage has quite regularly relied also on the growth of performance driving that storage.
VastData has achieved remarkable function in the growth of both those key categories.
- Performance – The leverage of 3D Crosspoint in the storage environment has granted for a much more robust handling of the Metadata in a proactive manner. The Crosspoint is in place mostly as a write cache. The environment also leans on newer and faster storage platforms, like NVMe for the JBOF (Just a bunch of Flash). The performance from within each JBOF is further enhanced by multiple either ethernet or Infiniband connections out from each JBOF into the server environment. Additionally, the compute nodes are essentially stateless, and can be updated to new versions of code by simply pushing out the next generation of container-based OS. It’s a very good model, and very easy to maintain consistency across the compute layer.
- Scalability – Unlike traditional monolithic storage arrays, where scalability may work within the rack, growing the environment in sort of blocks of arrays, the build in the Vast architecture disaggregates the storage from the compute, so that should more horsepower be required, that would be a server-side growth, while alternatively, the same reliance on storage growth requires additional JBOF, yet not necessarily compute.
As you can see, the architecture is designed around the idea that all data, regardless of its traditional location within the storage environment, including all levels of the pyramid can be handled by the varying levels of performance, and overlaid file-system types that can be approached by the Vast architecture. The Universal File System (UFS) is critical to the support of variable blocks in the storage infra, but also allows for one global namespace to be able to support variable block dedupe, and maintain that functionality within the compute layer rather than the storage layer.
Ultimately, I’m finding the architecture to be both modern and to address key limitations that have existed within Posix for a long time. To be fair, when Howard Marks (@DeepStorageNet) joins a startup as the tech evangelist, one must stand up and take notice.
So, in my opinion, particularly as a launch from Stealth, this was a particularly compelling presentation of a particularly interesting piece of storage tech. To gain some deeper knowledge than this posting, please get yourself over to the Tech Field Day youtube channel here. You will be entertained and educated.
One thought on “#SFD18 introduces us to VASTData”