3 ms·

Some of our GCP machines use far more than 80% CPU 24/7/365 on both preemptible and non-preemptible instances. We've yet to be asked to move to dedicated machin

Some of our GCP machines use far more than 80% CPU 24/7/365 on both preemptible and non-preemptible instances. We've yet to be asked to move to dedicated machines. It was pretty much the same when we were on AWS. I don't see how a cloud provider can complain about high CPU-usage if they bill by the machine and by the minute. Otherwise 20% of your bill is for idle CPUs.

tuetuopay 4y ago
It really depends how wide your impact is. When you have a lot of instances with this behaviour, you generate a lot of CPU steal for the neighbors on the same hypervisor (depending on the mitigations they put in place). If you happen to impact too many customers (many performance issue tickets), then the CSP will be more than happy to reach out to you to "upgrade" you. It really boils down to the business impact you have on them. Sure, you pay for the resource, but they are not guaranteed. And it is up to their discretion to de-prioritize your VMs on the HV, so that you're the last to get scheduled when the host's CPU is starved. This is the whole 'CPU credits' thing on AWS that they document in [0] where you can burst above the baseline. Interestingly, note that the baseline is in the 30% ballpark in their illustration on the bottom of the page :) [0]: https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/burstable-credits-baseline-concepts.html https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/burstabl...
- secondcoming 4y ago
  Fuck that. I'm paying for time on an agreed amount of vCPUs as specified in the instance templates. I'm using them 100%, I'd be a moron not to. It's also fraud because AWS get to charge me when I'm not scheduled, but they also charge the users who are scheduled on the contended CPU. They should bill by the CPU cycles I consume. Anyway, we've moved from AWS. Lucky escape.
- Hikikomori 4y ago
  This only applies to T type instances, that you likely wouldn't be using for something serious.
- gcbirzan 4y ago
  That aws things you linked is for a specific type of instances, for which they clearly state that you cannot run at 100 percent cpu all the time. It even says so in the first sentence > Traditional Amazon EC2 instance types provide fixed CPU resources, while burstable performance instances provide a baseline level of CPU utilization with the ability to burst CPU utilization above the baseline level. If the provider you are working for is doing this crap, then it sucks.

tuetuopay 4y ago

It really depends how wide your impact is. When you have a lot of instances with this behaviour, you generate a lot of CPU steal for the neighbors on the same hypervisor (depending on the mitigations they put in place). If you happen to impact too many customers (many performance issue tickets), then the CSP will be more than happy to reach out to you to "upgrade" you. It really boils down to the business impact you have on them. Sure, you pay for the resource, but they are not guaranteed. And it is up to their discretion to de-prioritize your VMs on the HV, so that you're the last to get scheduled when the host's CPU is starved. This is the whole 'CPU credits' thing on AWS that they document in [0] where you can burst above the baseline. Interestingly, note that the baseline is in the 30% ballpark in their illustration on the bottom of the page :) [0]: https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/burstable-credits-baseline-concepts.html https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/burstabl...

secondcoming 4y ago
Fuck that. I'm paying for time on an agreed amount of vCPUs as specified in the instance templates. I'm using them 100%, I'd be a moron not to. It's also fraud because AWS get to charge me when I'm not scheduled, but they also charge the users who are scheduled on the contended CPU. They should bill by the CPU cycles I consume. Anyway, we've moved from AWS. Lucky escape.
Hikikomori 4y ago
This only applies to T type instances, that you likely wouldn't be using for something serious.
gcbirzan 4y ago
That aws things you linked is for a specific type of instances, for which they clearly state that you cannot run at 100 percent cpu all the time. It even says so in the first sentence > Traditional Amazon EC2 instance types provide fixed CPU resources, while burstable performance instances provide a baseline level of CPU utilization with the ability to burst CPU utilization above the baseline level. If the provider you are working for is doing this crap, then it sucks.