Databases Reference
In-Depth Information
Solution
Remember that Exadata is merely an Oracle Database 11g Release 2 RAC database with up to eight nodes. So, all the
performance tuning techniques you learned in Chapter 16 apply to Exadata as well. In this recipe you will learn about
a few specifics relating to Exadata.
The storage for the database in Exadata comes from inside the rack. If you choose the RECO diskgroup as the fast
recovery area location, that also comes from the disks inside the Exadata and therefore there is a possibility of disk
contention. If you do see performance deteriorating for the RMAN output to the RECO diskgroup, you should check
the total I/O requests against that diskgroup. To that end, do the following:
1.
Log in as root to any one of the eight cell servers (not the Compute Node).
2.
Enter into CellCLI environment:
# cellcli
3.
Issue the following command in CellCLI:
CellCLI> list metriccurrent attributes metricObjectName, metricValue where name =
'GD_IO_RQ_W_LG' and metricObjectName like 'DWPRECO.*';
Here is the output:
DWPRECO_CD_00_dwhpcel01 3,933 IO requests
DWPRECO_CD_01_dwhpcel01 4,107 IO requests
DWPRECO_CD_02_dwhpcel01 4,106 IO requests
DWPRECO_CD_03_dwhpcel01 4,186 IO requests
DWPRECO_CD_04_dwhpcel01 3,947 IO requests
DWPRECO_CD_05_dwhpcel01 4,093 IO requests
DWPRECO_CD_06_dwhpcel01 3,765 IO requests
DWPRECO_CD_07_dwhpcel01 3,919 IO requests
DWPRECO_CD_08_dwhpcel01 4,067 IO requests
DWPRECO_CD_09_dwhpcel01 4,037 IO requests
DWPRECO_CD_10_dwhpcel01 4,007 IO requests
DWPRECO_CD_11_dwhpcel01 4,051 IO requests
The output shows that all the griddisks making up the RECO diskgroup on cell 01 have seen about the same
number of I/O requests. Had you seen a significant difference, you would have been concerned about a potential
issue on any griddisk showing a variance from the general trend.
Suppose you notice that the I/O requests coming to griddisk DWPRECO_CD_00_dwhpcel01 is very high
compared to the other griddisks. That by itself should not necessarily be a concern. You should examine whether the
upward swing in the I/O requests was consistent or just a blip. To know that, you should look at the history of this
specific metric on this griddisk by using the following CellCLI command (we have used a filter on the collection time
to limit the output to only the recent past):
CellCLI> list metrichistory attributes collectiontime, metricvalue where name = 'GD_IO_RQ_W_LG' and
metricObjectName = 'DWPRECO_CD_00_dwhpcel01' and collectionTime > '2012-09-08T00:00:00-04:00';
