This may seem like an obvious distinction, but a lot of coverage seems to gloss over the distinction between two aspects of support for grid or cloud computing:
- Making software run at all on many processors at once.
- Making it run fast on many processors.
In most cases, a good criterion for the latter would be near-linear horizontal scalability.
Let me illustrate by way of examples.
Elastra’s technology is all about opening and closing instances, managing/provisioning software, etc. I talked with them all through a multi-hour dinner and don’t even recall whether performance came up.
- GridGain seems to be all about performance. In this report on coming within 20% of linear scalability over 512 Amazon EC2 nodes, they seem to have thrown together a variety of custom scripts for starting/stopping instances or whatever. That simply doesn’t seem to be the focus of the core product.
- Pervasive Datarush — about which I know a lot more than I do about GridGain — is also all about performance. Indeed, it’s a library; all that administrivia about making things actually run is your job. (Or maybe the chip’s or hardware’s.)
On the other hand, if a vendor offers not grid/cloud support per se, but rather a product that happens to run in a massively parallel way, then it probably is handling both the challenges of basic operation and acceptable scalability. For example, Vertica told me it doesn’t need Elastra to do cloud database management, because it’s built similar capabilities into their own software. And the same is true of Vertica’s competitors; almost every serious vendor of data-warehouse-optimized database managers has its own integrated approach to handling massively parallel processing.
Related links
