Closed cy-zheng closed 5 years ago
One more question: Is there any plan to achieve data locality cost model, like Quincy?
Not at this time. We are currently working on the complex Gang Scheduling, including proportions, fairness etc. You are welcome to work on it, if you would like.
OK, thanks!
See here. https://kubernetes.io/docs/concepts/extend-kubernetes/poseidon-firmament-alternate-scheduler/
The link is broken now.
@anvithks
See here. https://kubernetes.io/docs/concepts/extend-kubernetes/poseidon-firmament-alternate-scheduler/
The link is broken now.
@deepak-vij The docs are based on Kubernetes version.
The documentation is available in the K8S v15 docs. Here you go https://v1-15.docs.kubernetes.io/docs/concepts/extend-kubernetes/poseidon-firmament-alternate-scheduler/
@anvithks , all the links on the Poseidon blog are broken now for some reason. Someone pointed it out. Can we fix it. thanks.
@anvithks, see the following link to the blog.
@anvithks, see the following link to the blog.
I will see if I can update these when I have some free time.
The links just need to be updated to the v1. 15 docs linked above.
@anvithks, thanks. Let me know when done so that I can let the community member know who wants to look at the docs. Thanks.
@anvithks, thanks. Let me know when done so that I can let the community member know who wants to look at the docs. Thanks.
The docs are available here now. No need to wait.
@chauncey-77, you can see the docs as @anvithks mentioned.
@chauncey-77, I will also publish the Google Doc for the CPU/Mem design on github.
It's very kind of you, thanks again!
@chauncey-77, you can see the docs as @anvithks mentioned.
@chauncey-77, I will also publish the Google Doc for the CPU/Mem design on github.
Hi, has the Coogle Doc for the CPU/Mem design been published on github? If it has, please share the link, thanks!
Sorry, I have been quite busy and ongoing holidays stuff. I am still trying to dig out information we put together at the time. Have not been able to find it yet. I will publish something when ready. In the meanwhile, code is quite self explanatory w.r.t. multi-dimensional CPU/Mem based cost model.
Sorry, I have been quite busy and ongoing holidays stuff. I am still trying to dig out information we put together at the time. Have not been able to find it yet. I will publish something when ready. In the meanwhile, code is quite self explanatory w.r.t. multi-dimensional CPU/Mem based cost model.
Yes I am reading the function in cpu_cost_model.h/cc, and I have figured out most of them. But at the same time I also have many questions. For example, I still can't understand the construction process of flow graph, i.e., how to generate the flow graph according to the CPU/Mem scheduling strategy. The cpu_cost_model involves some types of graph nodes, such as EquivClass, task, UnSceduledAgg, resource and sink, but I don’t figure out how these nodes form the flow graph. When comparing this graph construction function and those mentioned in the Firmament paper, I think they are different. So is there an example to understand the flow graph construction process like the Firmament paper do?
Yes, graph construction is different as we are dealing with complex multi-dimensional cost model. Firmament paper does not get into that. Give me some time, I will provide details.
Yes, graph construction is different as we are dealing with complex multi-dimensional cost model. Firmament paper does not get into that. Give me some time, I will provide details.
Oh this is awesome, I am very much looking forward to it.
See here. https://kubernetes.io/docs/concepts/extend-kubernetes/poseidon-firmament-alternate-scheduler/