4 ms·

Assuming grid computing ends up cheaper, any problem that doesn't depend on high availability. Why grid computing might be cheaper: people could rent out exces

by ikeboy 11y ago

Assuming grid computing ends up cheaper, any problem that doesn't depend on high availability. Why grid computing might be cheaper: people could rent out excess capability that would otherwise go wasted, so their marginal cost of computing is 0 or close to 0. The question is how much overhead would be, but building on a marginal cost of 0 is better for costs than any dedicated solution. It's not obvious that those considerations outweigh the cost efficiencies of centralization, but it's by no means obvious that they wouldn't or can't. Think the power grid as an example. If power wasn't fungible in close to real time (like cloud computing today, unless you specifically set it up that way) our prices would be more expensive, because less participants would be producing energy. The fact that anyone can, in theory, produce energy and sell it on a market which accepts the best bid improves the market for everyone. This seems similar.

roymurdock 11y ago
> any problem that doesn't depend on high availability Do you have any specific examples of resource-intensive computing problems (other than SETI@home) that do not depend on high availability? Seems a bit like a solution in search of a problem. I like your example of the power grid as an explanation for how owners of small amounts of computing power might interact with the larger system and have an impact on market prices. But if the analogy truly holds, it would be a good indication of how negligible rewards would be to smaller entities (individuals running computers or smartphones).
- ikeboy 11y ago
  Anything that would currently use AWS spot instances. Amazon created it, so we can assume there's a market for "cheaper, but lower availability". Google has something similar https://cloud.google.com/preemptible-vms/ https://cloud.google.com/preemptible-vms/ >I like your example of the power grid as an explanation for how owners of small amounts of computing power might interact with the larger system and have an impact on market prices. But if the analogy truly holds, it would be a good indication of how negligible rewards would be to smaller entities (individuals running computers or smartphones). I'm not thinking of individuals making a living off of their one computer. But surely there's a significant amount of computing power in datacenters that goes unused, or not optimally used. Now, they could manually partner with other datacenters to absorb excess capacity, and bargain a price individually. But if grid computing becomes a thing, all they'd have to do is hook up a bidder to their excess capacity. Admittedly, focusing on high capacity entities makes it harder to see micropayments as helpful. Even if the minimum transfer amount was $100, it would still be worth it for datacenter operators to trade on such a market.
  - eru 11y ago
    The smaller the minimum payment, the closer you can fit the area under the curve.
- jevinskie 11y ago
  BoilerGrid has been in use for years. I they have an automatic checkpoint system so your job/program can get evicted (either forced, like with a human logging into the node or via node failure), find a new node, and resume its progress. https://www.rcac.purdue.edu/compute/boilergrid/ https://www.rcac.purdue.edu/compute/boilergrid/

by ikeboy 11y ago

roymurdock 11y ago

> any problem that doesn't depend on high availability Do you have any specific examples of resource-intensive computing problems (other than SETI@home) that do not depend on high availability? Seems a bit like a solution in search of a problem. I like your example of the power grid as an explanation for how owners of small amounts of computing power might interact with the larger system and have an impact on market prices. But if the analogy truly holds, it would be a good indication of how negligible rewards would be to smaller entities (individuals running computers or smartphones).

ikeboy 11y ago
Anything that would currently use AWS spot instances. Amazon created it, so we can assume there's a market for "cheaper, but lower availability". Google has something similar https://cloud.google.com/preemptible-vms/ https://cloud.google.com/preemptible-vms/ >I like your example of the power grid as an explanation for how owners of small amounts of computing power might interact with the larger system and have an impact on market prices. But if the analogy truly holds, it would be a good indication of how negligible rewards would be to smaller entities (individuals running computers or smartphones). I'm not thinking of individuals making a living off of their one computer. But surely there's a significant amount of computing power in datacenters that goes unused, or not optimally used. Now, they could manually partner with other datacenters to absorb excess capacity, and bargain a price individually. But if grid computing becomes a thing, all they'd have to do is hook up a bidder to their excess capacity. Admittedly, focusing on high capacity entities makes it harder to see micropayments as helpful. Even if the minimum transfer amount was $100, it would still be worth it for datacenter operators to trade on such a market.
- eru 11y ago
  The smaller the minimum payment, the closer you can fit the area under the curve.
jevinskie 11y ago
BoilerGrid has been in use for years. I they have an automatic checkpoint system so your job/program can get evicted (either forced, like with a human logging into the node or via node failure), find a new node, and resume its progress. https://www.rcac.purdue.edu/compute/boilergrid/ https://www.rcac.purdue.edu/compute/boilergrid/