Yes, actually we are trying to compare and see what maximum TPS are we able to reach with both of these row by row and batch read/write test. And then afterwards, this figure may be compared with other databases etc with similar setups.
So wanted to understand from experts here, if this approach is fine? Or some other approach is advisable?
I agree to the point that , network will play a role in real world app, but here, we are mainly wanted to see the database capability, as network will always play a similar kind of role across all databases. Do you suggest some other approach to achieve this objective?
On Wed, 20 Dec, 2023, 2:42 am Peter J. Holzer, <hjp-pgsql@hjp.at> wrote:
On 2023-12-20 00:44:48 +0530, veem v wrote: > So at first, we need to populate the base tables with the necessary data (say > 100million rows) with required skewness using random functions to generate the > variation in the values of different data types. Then in case of row by row > write/read test , we can traverse in a cursor loop. and in case of batch write/ > insert , we need to traverse in a bulk collect loop. Something like below and > then this code can be wrapped into a procedure and passed to the pgbench and > executed from there. Please correct me if I'm wrong.
One important point to consider for benchmarks is that your benchmark has to be similar to the real application to be useful. If your real application runs on a different node and connects to the database over the network, a benchmark running within a stored procedure may not be very indicative of real performance.
hp
-- _ | Peter J. Holzer | Story must make more sense than reality. |_|_) | | | | | hjp@hjp.at | -- Charles Stross, "Creative writing __/ | http://www.hjp.at/ | challenge!"