This directory is a fork from https://github.com/Percona-Lab/sysbench-tpcc
The tests are scheduled by tpcc.lua, the usage is shown in runTests.sh, and the bash script could be adapted to run basically any test.
Some adaptations were made to run tests according to the needs of this project.
Most of the changes were made at tpcc.lua, including functions to calculate time delays outside of sysbench's core. There is also changes on tpcc_run.lua, as we wanted to isolate each client thread to it's own warehouse, so there would be no conflict between different client threads. This is visible at lines 77 and 270 of tpcc_run.lua, where the variable corresponding to the warehouse selected has the same value of the thread id. This way, the first client thread will operate on the warehouse 1, the second thread on the warehouse 2 and so on.
The tpcc_common.lua has changes too, since the original tpcc benchmark doesn't include q2. The new options implemented are:
-
thread_file A name under a directory. Each thread will create its own file storing the mean of all scheduling delays on its run. If the name given is "time_file", for example, the first thread will create the file "time_file_0", the second one will create "time_file_1" and so on.
-
include_q2 If set, sysbench will schedule q2 along new order and payment transactions. By default, the script will schedule batches of 10 queries and put q2 on a random position between then.
-
batch_size If include_q2 is not set, this flag is ignored. Otherwise, the scheduled batch has the given size. For example, if batch_size is 100, the q2 will be scheduled once for every 100 transactions.
-
batch_position If include_q2 is not set, this flag is ignored. Otherwise, q2 will be scheduled on the given position instead of randomly.
The main tests where done by the script runTests.sh. It is a great interface to run any needed test. Usually it is run on tmux session so it is not fixed to the ssh session. The results to the tests run until this point are under the results directory. The list of runned tests are the following:
-
generic_threads Every thread runs q2, new order and payment queries. Those tests were run before implementing the warehouse separation by client threads.
-
specialist_thread The only test made with one specialist thread running just Q2, the other threads ran new order and payment transactions
-
exclusiveWh The first tests run with exclusiveness for warehouses on each thread.
-
exclusiveNewtime Ran tests with exclusive warehouses and saving the scheduling times.
-
finalTests The same as exclusiveNewtime, but the results are separated by transaction type. These are the most complete tests done, there is a README.MD under that directory to explain the tests done and the scripts used to gather the statistics.