Mostly Moore's Law: costs of hardware keep falling. The initial Google search clusters were pretty small, a half dozen or so systems, though the Web was also a small fraction of its size.
The aspects of this I see being possible game-changers across the hosted SAAS space are of 1) autoconfiguring cheap systems, 2) better residential broadband, and 3) distributed tools which work very effectively even with tremendous decentralization.
On a per-user basis, the amount of compute resources allocated by even very large sites is relatively small. You can provision as much on your own. What providers do offer are engineering, robustness, reliability, and data redundancy.
Engineering can be provided through self-configuring systems -- a Debian base, set of packages, autoconfiguration, and a minimal set of things the user needs to worry about, and you're up and running. With SSD / flash drives, you buy yourself around much of the seek latency of spinning rust.
Robustness and reliability through a grid-provisioned service would ensure that the system as a whole is up even if individual nodes are offline. You'd also need some smarts against attacks, but if the system could communicate among nodes indicating what traffic is considered harmful it should be reasonably self-healing.
Data redundancy can be provided through sharding and replication schemes. Figuring out how to account for balance of data of a given node (if you want to load a lot you've got to offer a lot, and/or obtain credits somehow, perhaps through payment).
The one additional benefit DCs provide is for very rapid communications between nodes -- < 0.001s, sometimes better, latencies, rather than 0.01s - 0.1s between Internet sites (sometimes worse). This matters for multi-tier applications where web proxies, application engines, and databases need to talk to one another.
And if you do need some DC infrastructure, costs for that are falling (and capabilities increasing) such that much of it can be provided on a light budget (traffic costs are likely your biggest concern).
What I see is an erosion of the technological limitations which have made the present commercial model viable and/or necessary. Which suggests that it may be increasingly supplanted, especially for mature spaces.
The aspects of this I see being possible game-changers across the hosted SAAS space are of 1) autoconfiguring cheap systems, 2) better residential broadband, and 3) distributed tools which work very effectively even with tremendous decentralization.
On a per-user basis, the amount of compute resources allocated by even very large sites is relatively small. You can provision as much on your own. What providers do offer are engineering, robustness, reliability, and data redundancy.
Engineering can be provided through self-configuring systems -- a Debian base, set of packages, autoconfiguration, and a minimal set of things the user needs to worry about, and you're up and running. With SSD / flash drives, you buy yourself around much of the seek latency of spinning rust.
Robustness and reliability through a grid-provisioned service would ensure that the system as a whole is up even if individual nodes are offline. You'd also need some smarts against attacks, but if the system could communicate among nodes indicating what traffic is considered harmful it should be reasonably self-healing.
Data redundancy can be provided through sharding and replication schemes. Figuring out how to account for balance of data of a given node (if you want to load a lot you've got to offer a lot, and/or obtain credits somehow, perhaps through payment).
The one additional benefit DCs provide is for very rapid communications between nodes -- < 0.001s, sometimes better, latencies, rather than 0.01s - 0.1s between Internet sites (sometimes worse). This matters for multi-tier applications where web proxies, application engines, and databases need to talk to one another.
And if you do need some DC infrastructure, costs for that are falling (and capabilities increasing) such that much of it can be provided on a light budget (traffic costs are likely your biggest concern).
What I see is an erosion of the technological limitations which have made the present commercial model viable and/or necessary. Which suggests that it may be increasingly supplanted, especially for mature spaces.