I have about 5 different heterogenous linux dedicated server which can run Java applications.
Ideally, I would have one of the servers running a Java server like Netty acting as the “hub”. Clients would submit tasks to be run on the servers, and the first server would handle which tasks are to be run on available resources across different servers.
So if too many tasks are running already on one server, the first server “hub” would route the client tasks to be run on the next available server with free resources.
If the maximum task threshold is reached, a queue should be created. whenever resource is freed up, the next tasks in queue will be run.
Another part the hub should take care of is how often the task will be run. Once a month or weekly for example.
I would like to know what existing solutions there are (non-commercial, open source is best) for the above scenario. I do recall vaguely reading about a project on Netty server that does what I describe but I cannot recall the name of the project.
If you plan to use the netty framework, then you could the following.
For the ‘hub’ server create a pipeline in which you add a handler that spread the tasks to each underlying server. In that handler you could use blocking queues per server. If one server has too many tasks, they won’t be executed until the previous tasks are consumed by the underlying server. The part that will consume the tasks will be another handler that will act as a client to the actual underlying server. I guess you cannot forward the task directly to the servers. You need to use a client to request the task to be executed on a server.
Spreading the tasks with a round robin along the blocking queues in the “routing handler” will do.
Alternatively , you could have a look at MemoryAwareThreadPoolExecutor in Netty. And particularly at DefaultObjectSizeEstimator. You can limit how many request are processed by estimating the “weight” of the task.
Regards,
Tiberiu