Jeff Kubina wrote:
I read the following from the Accumulo manual on tablet merging <https://accumulo.apache.org/1.6/accumulo_user_manual.html#_merging_tablets>:Over time, a table can get very large, so large that it has hundreds of thousands of split points. Once there are enough tablets to spread a table across the entire cluster, additional splits may not improve performance, and may create unnecessary bookkeeping. So would the optimal number of tablets for a very large table be close to the total tservers times the total cores of the machine (or the worker threads the tservers are config to use--whichever is less)?
The general response normally given is "a few hundred of tablets" per server. I'm not sure if we really have a formula to compute this, though. I'm also curious how the things that are important to you affect this number (e.g. read latency, write throughput, read throughput, etc).
Would be neat to think through the math and create a little harness that actually test it out.
