sysbench の OLTP test statistics の write は DML 発行回数

sysbench で MySQL や PostgreSQL のベンチマークを取得することができるが、OLTP test statistics の write がDML発行回数かI/Oシステムコール発行回数のどちらか気になった。SQLインタフェースでアクセスしているので、DML発行回数だと想定していたが、sysbench のソースを確認したらやはり単純にDML発行回数のようだ。*1

つづいてベンチマークの取得です。

$ sysbench --test=/tmp/sysbench-0.5/sysbench/tests/db/oltp.lua \
> --db-driver=pgsql \
> --pgsql-user=sbuser --pgsql-password=sbpass --pgsql-db=sbdb \
> run
sysbench 0.5:  multi-threaded system evaluation benchmark

Running the test with following options:
Number of threads: 1
Random number generator seed is 0 and will be ignored


Initializing worker threads...

Threads started!

OLTP test statistics:
    queries performed:
        read:                            140000
        write:                           40000 ★
        other:                           20000
        total:                           200000
    transactions:                        10000  (96.53 per sec.)
    read/write requests:                 180000 (1737.48 per sec.)
    other operations:                    20000  (193.05 per sec.)
    ignored errors:                      0      (0.00 per sec.)
    reconnects:                          0      (0.00 per sec.)

General statistics:
    total time:                          103.5983s
    total number of events:              10000
    total time taken by event execution: 103.5635s
    response time:
         min:                                  4.04ms
         avg:                                 10.36ms
         max:                                 23.20ms
         approx.  95 percentile:              10.97ms

Threads fairness:
    events (avg/stddev):           10000.0000/0.00
    execution time (avg/stddev):   103.5635/0.00

KKIDA-GALAXY: sysbench 0.5 で PostgreSQL 9.5 のベンチマーク！環境構築編

• https://github.com/akopytov/sysbench/blob/master/src/db_driver.c

void db_report_cumulative(sb_stat_t *stat)
{
  sb_timer_t    exec_timer;
  sb_timer_t    fetch_timer;
 
  /* Use default stats handler if no drivers are used */
  if (!check_print_stats())
  {
    sb_report_cumulative(stat);
    return;
  }
 
  const double seconds = stat->time_interval;
  const uint64_t queries = stat->reads + stat->writes + stat->other; ★
 
  log_text(LOG_NOTICE, "SQL statistics:");
  log_text(LOG_NOTICE, "    queries performed:");
  log_text(LOG_NOTICE, "        read:                            %" PRIu64,
           stat->reads);
  log_text(LOG_NOTICE, "        write:                           %" PRIu64,
           stat->writes); ★
  log_text(LOG_NOTICE, "        other:                           %" PRIu64,
           stat->other);
  log_text(LOG_NOTICE, "        total:                           %" PRIu64,
           queries);
  log_text(LOG_NOTICE, "    transactions:                        %-6" PRIu64
           " (%.2f per sec.)", stat->events, stat->events / seconds);
  log_text(LOG_NOTICE, "    queries:                             %-6" PRIu64
           " (%.2f per sec.)", queries,
           queries / seconds);
  log_text(LOG_NOTICE, "    ignored errors:                      %-6" PRIu64
           " (%.2f per sec.)", stat->errors, stat->errors / seconds);
  log_text(LOG_NOTICE, "    reconnects:                          %-6" PRIu64
           " (%.2f per sec.)", stat->reconnects, stat->reconnects / seconds);

*1:Oracle Database のようにIOリクエスト回数をカウントしていてSQLインターフェースで情報取得可能ならI/Oシステムコール発行回数を取得することも可能だが