Bug 13510853 Linux: Slow DB startup with many files due to DBW time in skgfrgblkdev()
Affects:
| Product (Component) | Oracle Server (Rdbms) |
| Range of versions believed to be affected | (Not specified) |
| Versions confirmed as being affected |
|
| Platforms affected |
|
It is believed to be a regression in default behaviour thus:
Regression introduced in 11.2.0.2
Fixed:
Description
Some customers on the 11.2.0.2 release of Oracle on Linux x86-x64 may see
symptoms of slow database start up. Supporting diagnostics such as strace
show a high number of accesses on /proc/diskstats. The customer database
also typically will have a high number of data files associated with it.
The root cause of the problem is the introduction of some new internal
stats collection in 11.2.0.2 which, in certain larger storage
configurations exhibit suboptimal performance.
The fix provided in this bug implements a means for customers to disable
this code path by defining an event in their init.ora/spfile (prior to
database start up.
Rediscovery Notes:
If you experience symptoms of slow database start up, strace diagnostics
collected on the oracle processes such as a dbwr show a high number of
accesses on /proc/diskstats, and you have a large number of datafiles that
are part of your database, you may be encountering this issue.
Note that the new internal stats collection that is root cause of this issue
is only present in 11.2.0.2 or later releases and only on the Linux ports of
the Oracle Server.
Stacks of DBW process is likely to show lots of time in/under skgfrgblkdev
eg: Example stack:
skgfrgblkdev <- ksfdstblkini <- ksfdfstini <- ksfdopn1
<- ksfdopn <- kcfbid <- kcfida ...
Workaround
None other than reduce the number of DB datafile significantly
Note:
This fix is disabled by default in 11.2.0.4 onwards .
To avoid the skgfrgblkdev calls in 11.2.0.4 onwards set event 10298
level 3489660928 (0xd0000000)
Niciun comentariu:
Trimiteți un comentariu