Originally from the User Slack
@NafNaf**:** Hi , Can someone tell me how to determine the best value for concurrent_writes configuration setting?
@Felipe_Cardeneti_Mendes
@Felipe_Cardeneti_Mendes**:** This is a noop parameter present simply for Cassandra compatibility, same for concurrent_reads and con@Nafurrent_counter_writes
@Naf**:** So setting this, no effect in scylladb?
and one more question how scylladb handles@Felipe_Cardeneti_Mendesconcurrent reads/writes?
@Felipe_Cardeneti_Mendes**:** Correc@Naf, I think it might even scream invalid on logs.
@Naf**:** I just want to know do scyl@Felipe_Cardeneti_Mendesa has any option to manage read/write concurrency?
@Felipe_Cardeneti_Mendes**:** It depends on what you refer to read/write concurrency. To simply answer your question, each shard (CPU core) works in a shared-nothing model, thus we specialize in high throughput use cases.
The options you refer to (from Cassandra) are mostly related to how Cassandra issues disk IO. In our case, we rely on an userspace I/O scheduler whose settings are within /etc/scylla.d/io_properties.yaml and priority classes for quality of service.
That said, I suppose your question is whether we have any options to “limit” the concurrency of incoming user requests (irrespective of compactions and other internal tasks). And yes, we do. max_concurrent_requests_per_shard is one example, reader_concurrency_semaphore_cpu_concurrency is another where you can serialize reads waiting for CPU time (not disk IO)
If you are getting started with Scylla, start with the documentation recommended settings. The defaults are sane for most use cases and require very little (if any) tuning unlike Cassandra
@Naf**:** Thank you very much!