 |
|
Oracle Tips by Burleson |
Finding
the source of bandwidth bottlenecks
The first step in applying SSD to RAC is to locate
the root cause of the I/O contention. The most common cause of I/O
bandwidth saturation in Oracle RAC is a poorly performing I/O
subsystem. However, more detailed exploration is warranted to see what
specific data files contribute to the saturation of the storage. The
source of Oracle I/O saturationmay
be due to one of these causes:
-
Non-database processes read from the same device
as Oracle data files
-
Another database sharing the same file systems
(Oracle RAC)
-
A poorly tuned I/O subsystem (e.g. RAID5 for
high-update data files).
There are two approaches to I/O monitoring Oracle
RAC for disk bandwidth bottlenecks:
– Use OS tools, such as
iostatof vendor-based
disk monitors.
– Use an AWRor STATSPACK report to
monitor buffer busy waitsand high
access times for each instance.
As a review, the hallmark feature of RAC is the
ability for many Oracle instances to simultaneously read the Oracle
files (Figure 2.6). This complicates the process of locating high
concurrent access data files because the file I/O on each of the RAC
instances must be interrogated.
Because Oracle file I/O is measured at the
instance level, finding bandwidth bottlenecks may mean reading AWRreports on dozens of RAC nodes. To simplify the process
of identification, most Oracle professionals will measure I/O at the
disk level using native I/O monitors such as SAR and
iostat.
They would then look for disk enqueues, a condition where read/write
requests are waiting for access to the disk.
In the example below, a SAR command was issued to
locate the disk enqueues:
root > sar –d –f /var/adm/sa/sa16
In the SAR output, review the avque column
seeking high device backlogs:
SunOS prod1 5.6
Generic_105181-23 sun4u 05/16/01
01:00:00 device
%busy avque r+w/s blks/s avwait avserv
sd22 100 72.4 2100 2971 0.0
87.0
sd23 0 0.0 0 0 0.0 0.0
sd24 0 0.0 0 0 0.0 0.0
sd25 100 72.4 2100 2971 0.0
87.0
Because of the transient nature of disk enqueues,
many Oracle professionals also use disk vendor specific time-series
tools (e.g. the EMC Symmetrics console) to track disk enqueues
over time.
Another approach to locating bandwidth bottlenecks
in Oracle RAC is to interrogate each instance for buffer busy waits.
The next section examines this closer.
The above book excerpt is from:
Oracle RAC & Tuning with
Solid State Disk
Expert Secrets for High
Performance Clustered Grid Computing
ISBN
0-9761573-5-7
Donald K. Burleson & Mike Ault
http://www.rampant-books.com/book_2005_2_rac_ssd_tuning.htm
|
