I think that the situation is more complicated if we start looking at machine prices. Let me use some real data: 1) I have to use a db machine like gcloud n1-standard-16 ---> kubernetes cluster with 1 node for 500$/month 2) I have to use 9 web server like n1-standard-2 ---> kubernetes cluster with 9 nodes for 480$/month
So with about 1000$/month I have the configuration that currently supports the web traffic of my company. If I wanted to use a single cluster I should choose nodes like n1-standard-16. Wanting not to exceed the $1000 limit, I could create a cluster with 2 nodes. So I'll have: a node for db and a node for 9 (web) pod So the real question could be: in terms of performance, scalability and stability which is the better solution between: (9 nodes with 1 pod) vs (1 node with 9 pods) If two alternatives are comparable I could use a single cluster :) Il giorno martedì 12 dicembre 2017 23:00:10 UTC+1, David Rosenstrauch ha scritto: > On 2017-12-12 4:38 pm, Marco De Rosa wrote: > > The main reason is that the "web" cluster has hardware features > > different from the "db" cluster and I didn't find a way to have a > > cluster with for example one node better, in cpu and/or ram, than > > others. > > So 2 clusters to put in communication with the doubt that I have > > described above. > > The alternative could be create a single cluster with n nodes sized in > > such a way as to support web traffic and database work. > > So a situation where I have for example 4 nodes: in 3 nodes 6 web-pods > > plus the last node as pure db machine. > > But this solution is quite complicated in terms of how precisely to > > size the web pods, the db and the overall characteristics of the > > cluster.. > > So the idea to create two different clusters > > > FYI, this could probably be easily accomplished on a single cluster, > using node labels and node selectors. > > Let's say you had 2 types of nodes: machines with big disks, and > machines with lots of memory. Then let's say that you have 2 different > types of containers - one that runs a memory cache, and one that runs a > log file processing system. What you could do is label the nodes as, > say, either "type=hidisk" or "type=himem", as appropriate. And then you > could set a node selector on the caches to only run on nodes with > "type=himem", and a node selector on the log processors to only run on > nodes with "type=hidisk". > > HTH, > > DR -- You received this message because you are subscribed to the Google Groups "Kubernetes user discussion and Q&A" group. To unsubscribe from this group and stop receiving emails from it, send an email to kubernetes-users+unsubscr...@googlegroups.com. To post to this group, send email to kubernetes-users@googlegroups.com. Visit this group at https://groups.google.com/group/kubernetes-users. For more options, visit https://groups.google.com/d/optout.