Help me tuning this wait event:log file sync

hi, dbas: One of the database i manage face a serious wait event, log file sync. This is a big and busy oltp system, and using disk array of Sun T3 with raid-5. We are using Veritas QuickIO for datafile and redo log files. The pressure on the database is growing fast, and more and more the redo log becomes the bottleneck of the database. Here is some data showing the fact:

2002/05/03 Top 5 Wait Events ~~~~~~~~~~~~~~~~~ Wait % Total Event Waits
        Time (cs) Wt Time --------------------------------------------


        ------------ ------------ ------- log file sync 82,244 92,442

        33.68 db file sequential read 298,301 80,719 29.41 log file
        parallel write 81,849 36,989 13.48 db file parallel write 5,427
        33,615 12.25 control file parallel write 4,673 6,104 2.22


2002/06/07 Top 5 Wait Events ~~~~~~~~~~~~~~~~~ Wait % Total Event Waits
        Time (cs) Wt Time --------------------------------------------


        ------------ ------------ ------- log file sync 6,352,383

        15,785,313 40.09 db file sequential read 26,862,699 12,538,922
        31.85 log file parallel write 5,971,229 3,990,066 10.13 db file
        parallel write 290,479 3,164,391 8.04 db file scattered read
        1,749,137 814,981 2.07

2002/07/21 Top 5 Wait Events ~~~~~~~~~~~~~~~~~ Wait % Total Event Waits
        Time (cs) Wt Time --------------------------------------------


        ------------ ------------ ------- log file sync 2,207,609

        6,688,751 59.47 log file parallel write 2,044,977 1,385,379
        12.32 db file parallel write 103,155 1,203,077 10.70 db file
        sequential read 8,772,908 1,088,922 9.68 log buffer space 3,284
        222,604 1.98

2002/08/28 Top 5 Wait Events ~~~~~~~~~~~~~~~~~ Wait % Total Event
        Waits Time (cs) Wt Time
        -------------------------------------------- ------------
        ------------ ------- log file sync 2,247,585 20,529,779 63.71
        db file parallel write 441,052 4,377,899 13.59 log file
        parallel write 1,724,089 3,806,535 11.81 db file sequential
        read 8,854,525 2,044,020 6.34 enqueue 78,759 592,411 1.84

2002/09/04 Top 5 Wait Events ~~~~~~~~~~~~~~~~~ Wait % Total Event Waits
        Time (cs) Wt Time --------------------------------------------


        ------------ ------------ ------- log file sync 3,838,694

        13,158,371 63.28 db file sequential read 2,189,863 2,401,275
        11.55 log file parallel write 3,401,035 2,098,018 10.09 db file
        parallel write 97,086 1,503,608 7.23 enqueue 71,251 432,706 2.08

2002/10/11 Top 5 Wait Events ~~~~~~~~~~~~~~~~~ Wait % Total Event Waits
        Time (cs) Wt Time --------------------------------------------


        ------------ ------------ ------- log file sync 453,862

        9,679,513 68.11 db file sequential read 500,157 1,660,982 11.69
        db file parallel write 67,235 1,137,017 8.00 log file parallel
        write 163,180 1,075,201 7.57 log file sequential read 32,012
        254,327 1.79

As you can see, log file sync is a big bottleneck, and i have to solve this problem. I think the best solution maybe convert to dedicated raid1 or single disk with oracle multiplex, but this need to rebuild the whole disk array , need too much downtime, and with some risk. I tried to disable CachedQuick IO on the redo log file, but did not see performance gains. Veritas Said quick io is as fast as RAW, but i did not tested it before, i wonder whether move to raw for redo log will helps? It is difficult to modify the application to reduce the frequency of commit, for every transaction finishes, it have to commit.

Another strange question is that: from the iostat result and sar -d result, there is not much write to the disk array, maybe 2Mbytes/second, and it causes so much wait( average 10%time wait during normal workload). Maybe some misconfiguration in the disk array? Thanks for everyone reading and answering, please give me some suggestions

--
an oracle fan, 
an oracle beginner

Posted via dBforums
http://dbforums.com