kkaffes / voltdb-reproduction Goto Github PK

Reproduction of VoltDB's reduced performance with large record sizes.

This benchmark inserts records to a VoltDB table as fast as possible
and measures the throughput and latency of the insert operations.

Usage: ./build/bin/reproduction[options]
	-h: show this message
	-o <output log file path>: default reproduction_results.csv
	-s <DB server>: default 'localhost'
	-i <measurement interval>: default 2000 msec
	-t <total execution time>: default 10000 msec
	-m <message size>: default 2048

We observe a steep performance drop when the record length increases
from 8KB to 16KB:

./build/bin/reproduction -m 8192
Total transactions: 112750
Average throughput: 11142.4

./build/bin/reproduction -m 16384
Total transactions: 192
Average throughput: 18.9706

Setup:
1. Make sure that voltdb runs on localhost.
2. Build the code:
      a. mkdir build
      b. cd build
      c. cmake ..
      d. make
2. Set the "VOLT_HOME" and "VOLTDB_BIN" environment variables.
3. Insert the table: ./scripts/create_tables.sh
4. Create the stored procedure: ./scripts/update_voltdb_procedures.sh