Storage Performance Impact

Analytics statistic collection comes at some cost to overall performance. This should not be an issue if you understand what that cost will be, and how to minimize or avoid it.

Analytics statistics can be archived, meaning they will be a dataset that is continually read and saved to the Oracle ZFS Storage Appliance system disks in one-second summaries. This allows statistics to be viewed month by month, day by day, right down to second by second. Data is not discarded. If an Oracle ZFS Storage Appliance system has been running for two years, you can zoom down to by-second views for any time in the previous two years for your archived datasets. Depending on the type of statistic, this could present an issue with system disk usage.

You can monitor the growing sizes of the datasets, and destroy datasets that are growing too large. The system disks have compression enabled, so the sizes visible in the datasets view will be larger than the space consumed on disk after compression. For system disk usage and available space, see Viewing System Disks Status in Oracle ZFS Storage Appliance Customer Service Manual, Release OS8.8.x.

The following are example sizes taken from an Oracle ZFS Storage Appliance system that has been running for over four months.

Table 5-1 Sizes Taken from an Oracle ZFS Storage Appliance System Running for over Four Months

Category	Statistic	Span	Dataset Size*	Disk Consumed*
CPU	Percent utilization	130 days	127 MB	36 MB
Protocol	NFSv3 operations per second	130 days	127 MB	36 MB
Protocol	NFSv3 operations per second broken down by type of operation	130 days	209 MB	63 MB
CPU	Percent utilization broken down by CPU mode	130 days	431 MB	91 MB
Network	Device bytes per second broken down by device	130 days	402 MB	119 MB
Disk	I/O bytes per second broken down by disk	130 days	2.18 GB	833 MB
Disk	I/O operations per second broken down by latency	31 days	1.46 GB	515 MB

* These sizes will vary depending on your workload; they have been provided as a rough guideline.

It is worth noting that Oracle ZFS Storage Appliance has been intended to have 500 gigabyte mirrored system disks, most of which will be available to store datasets.

The factors that affect consumed disk space are:

• Type of statistic: Raw versus breakdowns

• For breakdowns: Number of breakdowns, and breakdown name length

• Activity rate

Monitor the size in Datasets. If a dataset is growing too large, and you want to stop it from growing but keep the historic data, use the suspend action.

Raw Statistics

Statistics that are a single value (sometimes written "as a raw statistic") will not consume much disk space for these reasons:

• Integer values consume a fixed and small amount of space

• The archives are compressed when saved, which significantly reduces the size for statistics that are mostly zero

Examples:

• CPU: Percent utilization

• Protocol: NFSv3 operations per second

Breakdowns

Statistics that have breakdowns can consume much more data, as shown in the previous table, because:

• Each breakdown is saved per second. For by-file and by-hostname breakdowns, the number of breakdowns per second may reach into the hundreds (how many different files or hosts had activity in a one-second summary), and all of which must be saved to disk.

• Breakdowns have dynamic names, which themselves can be long. You may only have ten active files in your breakdown by-file statistics, but each pathname could be dozens of characters in size. This might not sound like much, but the dataset will grow steadily when this data is saved every second.

Examples:

• CPU: Percent utilization broken down by CPU mode

• Protocol: NFSv3 operations per second broken down by type of operation

• Disk: I/O bytes per second broken down by disk

• Disk: I/O bytes per second broken down by latency

Exporting Statistics

At some point in time, you might want to archive statistics on a different server, either to free up disk space on Oracle ZFS Storage Appliance or for other purposes. For more information about exporting statistics or downloading statistic data in a comma-separated values (CSV) format, see Working with Analytics.