Hortonworks recommends the following guidelines for all Storm topologies.
Note | |
---|---|
These recommendations focus on guidelines for writing and debugging Storm topologies, rather than hardware tuning. Typically, most of the computation burden falls on the Supervisor and Worker nodes in a Storm cluster. The Nimbus node usually has a lighter load. For this reason, Hortonworks recommends that organizations save their hardware resources for the relatively burdened Supervisor and Worker nodes. |
Table 1.5. Storm Topology Guidelines
Guideline | Description |
---|---|
Read topology configuration parameters from a file. | Rather than hard coding configuration information in your Storm application, read the configuration parameters, including parallelism hints for specific components, from a file inside the |
Use a cache. | Use a cache to improve performance by eliminating unnecessary operations over the network, such as making frequent external service or lookup calls for reference data needed for processing. |
Tighten code in the | Every tuple is processed by the |
Perform benchmark testing to determine latencies. | Perform benchmark testing of the critical points in the network flow of your topology. Knowing the capacity of your data "pipes" provides a reliable standard for judging the performance of your topology and its individual components. |