4 ms·

But at Google you fall in to the first category, right? Where Google has a ton of servers sitting idly all the time, so it costs them nothing (or nearly nothing

But at Google you fall in to the first category, right? Where Google has a ton of servers sitting idly all the time, so it costs them nothing (or nearly nothing) to let you use a few thousand of them to do some testing. Or were you renting the server time from another company?

virgilp 6y ago
It was same when I worked at Adobe, really. And Adobe doesn't have servers sitting idly all the time. In the old days, getting approvals for a cluster, while feasible, would be a big overhead, take time, and then in 1 year it'd be already substandard. With cloud computing, I could easily spin up a huge cluster without asking anybody for approval, do some work on it, and turn it off after an hour. I could see how for small companies, "spinning up a VM" would be a thing to consider carefully - but for bigger ones (and especially for "big data" workloads), cloud computing is really a no-brainer.
- throwdbaaway 6y ago
  I think what zmmmmm had in mind is the mindset difference between unmetered usage and metered usage. For unmetered usage, imagine that due to some circumstances, you end up having exclusive access to 2 powerful physical machines where you could do anything you want. These machines will be idling 95% of the time, but when you need it, they are ready to serve you immediately, with zero friction. For metered usage, that would be public cloud, as you described.
  - username90 6y ago
    The workers I had at Google were definitely metered, just that the quota we have was very high so a thousand workers was no big deal as long as you turned them down after. I mean, if you give engineers cloud computing resources equal to a few percent of the cost of having said engineer then you can do this easily. And why not give them that? 20 engineers and $100k worth of experimental cloud quota is worth a lot more than 21 engineers and very limited cloud quota, at least if they are working with data pipelines and such.

virgilp 6y ago

It was same when I worked at Adobe, really. And Adobe doesn't have servers sitting idly all the time. In the old days, getting approvals for a cluster, while feasible, would be a big overhead, take time, and then in 1 year it'd be already substandard. With cloud computing, I could easily spin up a huge cluster without asking anybody for approval, do some work on it, and turn it off after an hour. I could see how for small companies, "spinning up a VM" would be a thing to consider carefully - but for bigger ones (and especially for "big data" workloads), cloud computing is really a no-brainer.

throwdbaaway 6y ago
I think what zmmmmm had in mind is the mindset difference between unmetered usage and metered usage. For unmetered usage, imagine that due to some circumstances, you end up having exclusive access to 2 powerful physical machines where you could do anything you want. These machines will be idling 95% of the time, but when you need it, they are ready to serve you immediately, with zero friction. For metered usage, that would be public cloud, as you described.
- username90 6y ago
  The workers I had at Google were definitely metered, just that the quota we have was very high so a thousand workers was no big deal as long as you turned them down after. I mean, if you give engineers cloud computing resources equal to a few percent of the cost of having said engineer then you can do this easily. And why not give them that? 20 engineers and $100k worth of experimental cloud quota is worth a lot more than 21 engineers and very limited cloud quota, at least if they are working with data pipelines and such.