Monitoring Exadata Storage Cells - Oracle Exadata Recipes: A Problem-Solution Approach - page 378

or for any given time from your metric history (in other words, metrichistory ). To list any metrics currently in an
abnormal state, run the following CellCLI command:

CellCLI> list metriccurrent where alertState != 'Normal'
CellCLI>

Here, we see no evidence of abnormal metric conditions. Next, query your metrichistory object to list all
abnormal metrics:

CellCLI> list metrichistory where alertState != 'Normal'
CellCLI>

■ the alertState of a metric is governed by seeded or customized metric thresholds . to learn more about
metric thresholds, please see recipe 13-4.

Note

How It Works
Exadata Storage Server metrics are recorded observations of important runtime properties or internal
instrumentation values of the storage cell and its components. Exadata's CELLSRV (Cell Services) and MS
(Management Services) processes work in tandem to track and store metric information.

• Exadata's Cell Services software, CELLSRV, is instrumented to track several hundred metrics

and stores these in storage server memory. When you execute a list metriccurrent
command from CellCLI, you are accessing the in-memory storage server metrics.

• Every hour, the MS process writes the in-memory metric information to disk and flushes

current metrics from memory. Once written to disk, the data is available through the CellCLI
metrichistory object.
MS retains seven days of metrics on disk and automatically ages out information older than

•

seven days.

The three main objects that contain metric information are metricdefinition , metriccurrent , and
metrichistory . In the Problem section of this recipe, we used the term “data model”; the metric hierarchy is not
actually stored in a relational model, but the attributes of each object are related.
As mentioned, there are a number of metric object types, each of which is associated with a specific component
of the storage server infrastructure. You can report current or past metrics for each metric and each classification of
objectType using the same list metriccurrent and list metrichistory commands. In the following sections, we
provide some additional detail and monitoring recommendations per objectType :

Cell Disks (objectType = CELLDISK)
Cell disks metrics are broken down into two main categories: I/O requests and I/O throughput. These are further provided
for both small I/O and large I/O requests, and each of these is also classified based on a cumulative and rate metrics.

• Look for a relatively smooth distribution of I/O requests, MB per second, and I/O latencies

across each cell disk and Exadata storage cell.

• Look for relatively low I/O latencies; under 10ms is generally considered an ideal target for

small I/Os, but larger I/O operations can have higher values.

• Monitor your I/O load using the

CD_IO_LOAD metric and watch for high utilization.