Re: I/O issues on DB 11g
From: David Fitzjarrell <oratune_at_yahoo.com>
Date: Wed, 23 Apr 2014 13:54:12 -0700 (PDT)
Message-ID: <1398286452.21380.YahooMailNeo_at_web124704.mail.ne1.yahoo.com>
Since you have both the controlfiles and the redo logs on the same disk as the database that is likely impacting the I/O operations for the database; db files are random-access, scattered read and write files and the controlfiles/redo logs are sequential read/write files. Think about this for a moment; you're mixing write patterns and that can slow down the datafile reads/writes as you wait for the disk heads to finish a redo or controlfile entry; the opposite is also true as controlfile/redo writes wait on datafifle processing. It's not an ideal situation. You might consider having a disk solely for controlfiles and redo logs. David Fitzjarrell Primary author, "Oracle Exadata Survival Guide" On Wednesday, April 23, 2014 2:40 PM, David Ramírez Reyes <dramirezr_at_gmail.com> wrote: Hi All, This is the environment. Windows 2008 R1 Standard, Oracle DB 11g Standard R2, 8 cpu's, 16 GB of physical memory, 3 disk drives (1 for the OS, 1 for db files, 1 for backups). Now the problem: Since we went live with the 11g system about 1 year ago (we used to be on a very old and horrible 8i -don't ask why-), we have been receiving Email alerts about Disk Utilization; at the beginning I thought it should be a bug of the R2 version as I wrongly understood it was referring about filesystem space, which is not a problem. After several months of 5 or 8 daily mails, I decided to look at it on detail and check what was necessary to drop off that "false alarm". After Goggling, I realized that the alarm is not related to disk space, but I/O reads, as we have 3 db's on the same disk drive, each of them with 20 db files (the biggest DB has datafiles of about 6 GB, the smallest about 2 GB). The problem is not really "critical" now because general performance is "good" (we have more than a year with it!), but that of course does not mean it has to keep on with those problems (and that alarm is starting causing me headaches also!). The first two things I though were increasing the PGA size in order to reduce Virtual Memory usage (and, I/O as consequence) and add 2 more disk drives to split the db files of each db into a single and dedicated filesystem; I was also thinking about tuning some high I/O queries, but don't think the difference could be huge... Any ideas or suggestions? Thanks David Ramírez
Date: Wed, 23 Apr 2014 13:54:12 -0700 (PDT)
Message-ID: <1398286452.21380.YahooMailNeo_at_web124704.mail.ne1.yahoo.com>
Since you have both the controlfiles and the redo logs on the same disk as the database that is likely impacting the I/O operations for the database; db files are random-access, scattered read and write files and the controlfiles/redo logs are sequential read/write files. Think about this for a moment; you're mixing write patterns and that can slow down the datafile reads/writes as you wait for the disk heads to finish a redo or controlfile entry; the opposite is also true as controlfile/redo writes wait on datafifle processing. It's not an ideal situation. You might consider having a disk solely for controlfiles and redo logs. David Fitzjarrell Primary author, "Oracle Exadata Survival Guide" On Wednesday, April 23, 2014 2:40 PM, David Ramírez Reyes <dramirezr_at_gmail.com> wrote: Hi All, This is the environment. Windows 2008 R1 Standard, Oracle DB 11g Standard R2, 8 cpu's, 16 GB of physical memory, 3 disk drives (1 for the OS, 1 for db files, 1 for backups). Now the problem: Since we went live with the 11g system about 1 year ago (we used to be on a very old and horrible 8i -don't ask why-), we have been receiving Email alerts about Disk Utilization; at the beginning I thought it should be a bug of the R2 version as I wrongly understood it was referring about filesystem space, which is not a problem. After several months of 5 or 8 daily mails, I decided to look at it on detail and check what was necessary to drop off that "false alarm". After Goggling, I realized that the alarm is not related to disk space, but I/O reads, as we have 3 db's on the same disk drive, each of them with 20 db files (the biggest DB has datafiles of about 6 GB, the smallest about 2 GB). The problem is not really "critical" now because general performance is "good" (we have more than a year with it!), but that of course does not mean it has to keep on with those problems (and that alarm is starting causing me headaches also!). The first two things I though were increasing the PGA size in order to reduce Virtual Memory usage (and, I/O as consequence) and add 2 more disk drives to split the db files of each db into a single and dedicated filesystem; I was also thinking about tuning some high I/O queries, but don't think the difference could be huge... Any ideas or suggestions? Thanks David Ramírez
-- http://www.freelists.org/webpage/oracle-lReceived on Wed Apr 23 2014 - 22:54:12 CEST