Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss

log file sync vs log file parallel write

14 views
Skip to first unread message

HansP

unread,
Oct 3, 2008, 6:31:48 AM10/3/08
to
Hello all,

While investigating performance problems in a database I discovered
that the 'log file sync' wait event is almost always in the top 5.
(I am looking at the statspack datat)

Calculating the ratio of the average wait time of the 'log file sync'
and the 'log file parallel write' gives a ratio of 10.

Does someone know why this may happen?

Regards Hans-Peter

sybrandb

unread,
Oct 3, 2008, 7:22:11 AM10/3/08
to

First of all,

'Log file sync' is associated with commit.
If you see this event in the top 5, you are committing way too often
(maybe for every individual record) and it is
developer beating time!!! They don't know what a transaction is, and
they need to be whipped for that.
Secondly 'log file parallel write' obviously deals with flushing out
the redo log buffer to disk. This means your disks are too slow, or
you have an I/O bottleneck.
Please note there is plethora of resources on 'log file sync' online,
particularly in this very forum.
Please force yourself being less lazy and use Google, your friend, to
search this and other forums, instead of asking this boring FAQ AGAIN.
The basic problem with many newbie DBAs is not they are short of
knowledge, the basic problem is they are too lazy, and want to be
spoon fed.
That won't bring you anywhere.

--
Sybrand Bakker
Senior Oracle DBA

Steve Howard

unread,
Oct 3, 2008, 8:09:14 AM10/3/08
to

Hi Hans-Peter,

When you say...

>>average wait time of the 'log file sync' and the 'log file parallel write' gives a ratio of 10.<<

...do you mean each log file parallel write takes ten times longer on
average than each log file sync?

Also, what version are you on? There was a bug in 9i (I forget which)
where "log file parallel write" was "under-reported". Jonathan Lewis
had a writeup on it.

Regards,

Steve

HansP

unread,
Oct 3, 2008, 8:13:40 AM10/3/08
to
First of all I am not a newbie.
I know what the meaning of the events are.

I googled and read books that are available on performance tuning.

The only thing I did not understand is that the average write to the
redo logs is about 10ms.
But why should the log file sync be 10 times as high.

In the mean time I continued reading and came across a not of Jonathan
Lewis about bug 2669566
So I think I may be hitting this bug.

You should not judge that fast.

HansP

unread,
Oct 3, 2008, 8:26:03 AM10/3/08
to
Hello Steve,

I am on 9.2.0.8.
Yes the log file sync is on average 10 times the average log file
parallel write .

I already came across the bug (2669566) on Jonathan Lewis is site.
So probably this bug applies.

regards Hans-Peter

0 new messages