5 ms·

This is a great (read:inexpensive) resource for big data modelling if someone is willing to build for it. By choosing to be the customer of least concern and de

This is a great (read:inexpensive) resource for big data modelling if someone is willing to build for it. By choosing to be the customer of least concern and designing software that is able to handle being bounced, you are getting a terrific deal, assuming your work isn't particularly time sensitive, such as in a bioinformatics research lab. I wish they put the minimum time you will be granted cycles though, as that information seems like an important design constraint when choosing the size of data chunks you design for.

LeifCarrotson 9y ago
> Compute Engine may shut them down after providing you a 30-second warning, and you can use them for a maximum of 24 hours. That's a good start...looks like the model is "save your work often". I'm sure you could run a few instances for a while and model the distribution of time granted.
- viridian 9y ago
  Right, my concern is more the event of them axing work 32 seconds in or something, which could be a problem for some data analytics tools I've seen, that need a good 70 seconds or so to reach a steady state.
  - jkaplowitz 9y ago
    Quoting their docs: "Compute Engine doesn't bill for instances preempted in the first 10 minutes," so that scenario is harmless in most application designs where preemptible VMs would make sense anyway. Disclosure: I used to work at Google on GCE, but not directly on preemptible VMs or this billing policy. I don't currently work or speak for Google.
gcr 9y ago
This is exactly how the AWS spot market works. If you don't mind getting bumped off, you can cut GPU costs to 30% of the usual.
candiodari 9y ago
Cloud is extremely expensive compared to your own hardware, especially for things like modelling (because even this price essentially is a great machine for every 4 months of compute, ignoring power). Plus the experience of doing modelling on a machine sitting under your desk is far superior to doing it remotely. Once you have a good pipeline and need to deploy it, sure. Then we can talk.

LeifCarrotson 9y ago

> Compute Engine may shut them down after providing you a 30-second warning, and you can use them for a maximum of 24 hours. That's a good start...looks like the model is "save your work often". I'm sure you could run a few instances for a while and model the distribution of time granted.

viridian 9y ago
Right, my concern is more the event of them axing work 32 seconds in or something, which could be a problem for some data analytics tools I've seen, that need a good 70 seconds or so to reach a steady state.
- jkaplowitz 9y ago
  Quoting their docs: "Compute Engine doesn't bill for instances preempted in the first 10 minutes," so that scenario is harmless in most application designs where preemptible VMs would make sense anyway. Disclosure: I used to work at Google on GCE, but not directly on preemptible VMs or this billing policy. I don't currently work or speak for Google.

gcr 9y ago

This is exactly how the AWS spot market works. If you don't mind getting bumped off, you can cut GPU costs to 30% of the usual.

candiodari 9y ago

Cloud is extremely expensive compared to your own hardware, especially for things like modelling (because even this price essentially is a great machine for every 4 months of compute, ignoring power). Plus the experience of doing modelling on a machine sitting under your desk is far superior to doing it remotely. Once you have a good pipeline and need to deploy it, sure. Then we can talk.