積読日記

■ [software] Firefox 3

早速入れた. いくつか対応していないアドオンがあったので代替を探してきた.

• FireGestures: AllinOneGestures から乗り換えた.
• Foxmarks: google browser sync が 3 に対応しないようなので乗り換えた.

■ [book] 野獣の都 〔永遠の戦士 ケイン〕, Michael Moorcock (著), 矢野 徹 (訳), ハヤカワ文庫 SF

読み終わった. 新シリーズよりも素朴で読みやすかった.

■ [programming] NUMA

NUMA (Non-Uniform Memory Access) の資料をまとめた. 手持ちの環境では NUMA の node が一つしかなかった. これでも NUMA の効果は実感できるのかなと思って AMD のデモを動かしてみた.

Allocate one big chunk of memory
  and set loose per-thread affinity before touching memory:
done loop= 8, ptr=08FE2000, mask= 3, thread=1, MB/sec=2754, 581 msec.
done loop= 9, ptr=09F24400, mask= 3, thread=1, MB/sec=2685, 596 msec.
done loop= 0, ptr=015D0000, mask= 3, thread=0, MB/sec=1113, 1437 msec.
done loop=10, ptr=0AE66800, mask= 3, thread=1, MB/sec=2812, 569 msec.
done loop=11, ptr=0BDA8C00, mask= 3, thread=1, MB/sec=2410, 664 msec.
done loop= 1, ptr=02512400, mask= 3, thread=0, MB/sec=1101, 1453 msec.
done loop=12, ptr=0CCEB000, mask= 3, thread=1, MB/sec=2619, 611 msec.
done loop=13, ptr=0DC2D400, mask= 3, thread=1, MB/sec=2645, 605 msec.
done loop=14, ptr=0EB6F800, mask= 3, thread=1, MB/sec=2740, 584 msec.
done loop= 2, ptr=03454800, mask= 3, thread=0, MB/sec=1085, 1474 msec.
done loop=15, ptr=0FAB1C00, mask= 3, thread=1, MB/sec=2730, 586 msec.
done loop= 3, ptr=04396C00, mask= 3, thread=0, MB/sec=2083, 768 msec.
done loop= 4, ptr=052D9000, mask= 3, thread=0, MB/sec=3893, 411 msec.
done loop= 5, ptr=0621B400, mask= 3, thread=0, MB/sec=4061, 394 msec.
done loop= 6, ptr=0715D800, mask= 3, thread=0, MB/sec=3791, 422 msec.
done loop= 7, ptr=0809FC00, mask= 3, thread=0, MB/sec=4134, 387 msec.
Done.  Total time = 6.98

Access the data with per-thread allocated memory
done loop= 8, ptr=02520000, mask= 3, thread=1, MB/sec=2703, 592 msec.
done loop= 9, ptr=02520000, mask= 3, thread=1, MB/sec=3206, 499 msec.
done loop= 0, ptr=015D0000, mask= 3, thread=0, MB/sec=1171, 1366 msec.
done loop=10, ptr=02520000, mask= 3, thread=1, MB/sec=2899, 552 msec.
done loop=11, ptr=02520000, mask= 3, thread=1, MB/sec=2947, 543 msec.
done loop= 1, ptr=015D0000, mask= 3, thread=0, MB/sec=1210, 1322 msec.
done loop=12, ptr=02520000, mask= 3, thread=1, MB/sec=3113, 514 msec.
done loop=13, ptr=02520000, mask= 3, thread=1, MB/sec=2952, 542 msec.
done loop=14, ptr=02520000, mask= 3, thread=1, MB/sec=3048, 525 msec.
done loop= 2, ptr=015D0000, mask= 3, thread=0, MB/sec=1198, 1336 msec.
done loop=15, ptr=02520000, mask= 3, thread=1, MB/sec=2980, 537 msec.
done loop= 3, ptr=015D0000, mask= 3, thread=0, MB/sec=1896, 844 msec.
done loop= 4, ptr=015D0000, mask= 3, thread=0, MB/sec=3636, 440 msec.
done loop= 5, ptr=015D0000, mask= 3, thread=0, MB/sec=4113, 389 msec.
done loop= 6, ptr=015D0000, mask= 3, thread=0, MB/sec=3563, 449 msec.
done loop= 7, ptr=015D0000, mask= 3, thread=0, MB/sec=4082, 392 msec.
Done.  Total time = 6.79

Use the standard Windows thread APIs instead, and set affinity:
done loop= 0, ptr=11C50000, mask= 3, thread=0, MB/sec=1606, 996 msec.
done loop= 1, ptr=10D00000, mask= 3, thread=1, MB/sec=2716, 589 msec.
done loop= 2, ptr=11C50000, mask= 3, thread=0, MB/sec=1641, 975 msec.
done loop= 3, ptr=10D00000, mask= 3, thread=1, MB/sec=2920, 548 msec.
done loop= 4, ptr=11C50000, mask= 3, thread=0, MB/sec=1700, 941 msec.
done loop= 5, ptr=10D00000, mask= 3, thread=1, MB/sec=2773, 577 msec.
done loop= 6, ptr=11C50000, mask= 3, thread=0, MB/sec=1598, 1001 msec.
done loop= 7, ptr=10D00000, mask= 3, thread=1, MB/sec=2837, 564 msec.
done loop= 8, ptr=11C50000, mask= 3, thread=0, MB/sec=1602, 999 msec.
done loop= 9, ptr=10D00000, mask= 3, thread=1, MB/sec=2610, 613 msec.
done loop=10, ptr=11C50000, mask= 3, thread=0, MB/sec=1610, 994 msec.
done loop=11, ptr=10D00000, mask= 3, thread=1, MB/sec=2787, 574 msec.
done loop=12, ptr=11C50000, mask= 3, thread=0, MB/sec=1576, 1015 msec.
done loop=13, ptr=10D00000, mask= 3, thread=1, MB/sec=2735, 585 msec.
done loop=14, ptr=11C50000, mask= 3, thread=0, MB/sec=1653, 968 msec.
done loop=15, ptr=10D00000, mask= 3, thread=1, MB/sec=2925, 547 msec.
Done.  Total time = 8.06