Hello, I am wondering if it is safe to assume that
specifying cio mount option is safe with PostgreSQL.
As far as I understand the CIO (AIX Concurrent I/O) means that
filesystem does not serialize access to file blocks. In other
words multiple threads can simultaneously read and write
the file block, which means that it is possible that reader
reads stale data. Now, if database enforces its own serialization
(and as far as I can tell, bufmgr does it exactly), this option
should be totally safe to use, probably giving boost since
kernel has less overhead while multiple processes access
the database file.
Is my thinking correct?
pgsql-performance by date
|Next:||From: Koichi Suzuki||Date: 2007-06-25 10:06:07|
|Subject: Re: PITR Backups|
|Previous:||From: Mikko Partio||Date: 2007-06-25 07:50:48|
|Subject: Re: Slow indexscan|