This overview describes the max latency/required throughput between the DMAs, between Cassandra nodes and between Elastic nodes.
However, what’s the required IOPS (IO operations per second)/disk throughput for DMAs, Cassandra nodes and Elastic nodes on their own? How fast should each of these servers be able to read/write to disk?
Hi Jochen, The HW requirements are available in the following article: https://community.dataminer.services/documentation/dataminer-system-requirements/. In there you will find a section for Cassandra nodes. The requirement is to have an SSD or mechanical drive for your OS and a separate SSD drive for the data tables of Cassandra. We request a separate disk for the Cassandra data and OS so that those two wouldn’t influence one another. Putting this in IOPS can be tricky as IOPS heavily depend on the data blocks you write to disk. Throughput (M/sB) = IOPS * Blocksize. In most cases you are limited by the link towards your disk; SATA II is for example up to 3Gb/s and SATA III is 6Gb/s. In case you are going to a virtualized environment, you will need to perform benchmark tests to evaluate if you are not influenced by other machines using the same resources (disk in this case).
Thanks for clarification what the difference is between IOPS and throughput. Still it’s not clear what the necessary throughput is between Cassandra node an its disk.