Using a unified architecture is a convenient way to learn and expand IoTDB indeed. If so, we should consider the extra resource cost in edge environment, which may need more test in later versions. For example, a 4G4C docker container can support X timeseries and Y QPS in a single process architecture. If it migrates to a 2 processes one, could it support the performance requirements above as well? And what's the best practice of the usage ratio between conifg node of the total limited available resource, 2:2 or 1:3, or any others?
> 在 2022年10月30日，13:07，HW-Chao Wang <[email protected]> 写道： > > +1，i support 1c1d for standalone verson，this is easy to learning for > users. and then cluster version will be a trend. > > > > ---Original--- > From: "Jialin Qiao"<[email protected]> > Date: Sun, Oct 30, 2022 10:29 AM > To: "dev"<[email protected]>; > Subject: About the standalone version > > > Hi, > > In this new cluster version, we separate the IoTDB service to two > process: ConfigNode and DataNode, which could be deployed on multiple > machines and one machine. > > So if users want to use a standalone IoTDB, they can deploy a 1C1D > architecture on one machine. > > We are investigating if we need to support a one-process(combine > ConfigNode and DataNode) architecture for the standalone version. > which does not see much significant but introduces some learning > difficulties to users from the product complexity perspective. > > In this case, I'd like to only release the ConfigNode and DataNode > Cluster, without another special standalone, and remove the > start-server.sh, start-new-server.sh. > > Instead, we could introduce a start-all.sh, which start one ConfigNode > and one DataNode by default. And this could also be used to start the > whole cluster. > > What do you think? > > Thanks, > ————————————————— > Jialin Qiao > Apache IoTDB PMC
