Vioperf

The vioperf utility quickly tests the performance of your host's input and output subsystem. The utility performs the following tests:

The utility verifies that the host reads the same bytes that it wrote and prints its output to STDOUT. The utility also logs the output to a JSON formatted file.

For data in HDFS, the utility tests reads but not writes.

Syntax

vioperf [--help] [--duration=<INTERVAL>] [--log-interval=<INTERVAL>] 
  [--log-file=<FILE>] [--condense-log] [--thread-count=<N>] [--max-buffer-size=<SIZE>]
  [--preserve-files] [--disable-crc] [--disable-direct-io] [--debug]
  [<DIR>*]

Minimum and Recommended I/O Performance

For example, the I/O rate for a node with 2 hyper-threaded six-core CPUs (12 physical cores) is 240 MB/s required minimum, 480 MB/s recommended.

Options

Option Description
--help

Prints a help message and exits.

--duration

The length of time vioprobe runs performance tests. The default is 5 minutes. Specify the interval in seconds, minutes, or hours with any of these suffixes:

  • Seconds: s, sec, secs, second, seconds. Example: --duration=60sec
  • Minutes: m, min, mins, minute, minutes. Example: --duration=10min
  • Hours: h, hr, hrs, hour, hours. Example: --duration=1hrs
--log-interval

The interval at which the log file reports summary information. The default interval is 10 seconds. This option uses the same interval notation as --duration.

--log-file

The path and name where log file contents are written, in JSON. If not specified, then vioperf creates a file named resultsdate-time.JSON in the current directory.

--condense-log

Directs vioperf to write the log file contents in condensed format, one JSON entry per line, rather than as indented JSON syntax.

--thread-count=<N>

The number of execution threads to use. By default, vioperf uses all threads available on the host machine.

--max-buffer-size=<SIZE>

The maximum size of the in-memory buffer to use for reads or writes. Specify the units with any of these suffixes:

  • Bytes: b, byte, bytes.
  • Kilobytes: k, kb, kilobyte, kilobytes.
  • Megabytes: m, mb, megabyte, megabytes.
  • Gigabytes: g, gb, gigabyte, gigabytes.
--preserve-files

Directs vioperf to keep the files it writes. This parameter is ignored for HDFS tests, which are read-only. Inspecting the files can help diagnose write-related failures.

--disable-crc

Directs vioperf to ignore CRC checksums when validating writes. Verifying checksums can add overhead, particularly when running vioperf on slower processors. This parameter is ignored for HDFS tests.

--disable-direct-io

When reading from or writing to a local file system, vioperf goes directly to disk by default, bypassing the operating system's page cache. Using direct I/O allows vioperf to measure performance quickly without having to fill the cache.

Disabling this behavior can produce more realistic performance results but slows down the operation of vioperf.

--debug

Directs vioperf to report verbose error messages.

<DIR>

Zero or more directories to test. If you do not specify a directory, vioperf tests the current directory. To test the performance of each disk, specify different directories mounted on different disks.

To test reads from a directory on HDFS:

  • Use a URL in the hdfs scheme that points to a single directory (not a path) containing files at least 10MB in size. For best results, use 10GB files and verify that there is at least one file per vioperf thread.
  • If you do not specify a host and port, set the HADOOP_CONF_DIR environment variable to a path including the Hadoop configuration files. This value is the same value that you use for the HadoopConfDir configuration parameter in Vertica. For more information see Configuring the hdfs Scheme.
  • If the HDFS cluster uses Kerberos, set the HADOOP_USER_NAME environment variable to a Kerberos principal.

Returns

The utility returns the following information:

Heading Description
test

The test being run (Write, ReWrite, Read, or Skip Read)

directory
The directory in which the test is being run.
counter name

The counter type of the test being run. Can be either MB/s or Seeks per second.

counter value

The value of the counter in MB/s or Seeks per second across all threads. This measurement represents the bandwidth at the exact time of measurement. Contrast with counter value (avg).

counter value (10 sec avg)

The average amount of data in MB/s, or the average number of Seeks per second, for the test being run in the duration specified with --log-interval. The default interval is 10 seconds. The counter value (avg) is the average bandwidth since the last log message, across all threads.

counter value/core
The counter value divided by the number of cores.
counter value/core (10 sec avg)
The counter value (10 sec avg) divided by the number of cores.
thread count

The number of threads used to run the test.

%CPU

The available CPU percentage used during this test.

%IO Wait

The CPU percentage in I/O Wait state during this test. I/O wait state is the time working processes are blocked while waiting for I/O operations to complete.

elapsed time

The amount of time taken for a particular test. If you run the test multiple times, elapsed time increases the next time the test is run.

remaining time

The time remaining until the next test. Based on the --duration option, each of the tests is run at least once. If the test set is run multiple times, then remaining time is how much longer the test will run. The remaining time value is cumulative. Its total is added to elapsed time each time the same test is run again.

Example

Invoking vioperf from a terminal outputs the following message and sample results:

[dbadmin@v_vmart_node0001 ~]$ /opt/vertica/bin/vioperf --duration=60s
The minimum required I/O is 20 MB/s read and write per physical processor core on each node, in full duplex 
i.e. reading and writing at this rate simultaneously, concurrently on all nodes of the cluster. 
The recommended I/O is 40 MB/s per physical core on each node.
For example, the I/O rate for a server node with 2 hyper-threaded six-core CPUs is 240 MB/s required minimum, 480 MB/s recommended.

Using direct io (buffer size=1048576, alignment=512) for directory "/home/dbadmin"

test     | directory     | counter name         | counter value | counter value (10 sec avg) | counter value/core  | counter value/core (10 sec avg) | thread count  | %CPU  | %IO Wait  | elapsed time (s)| remaining time (s)
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Write    | /home/dbadmin | MB/s                 | 420           | 420                             | 210                | 210                        | 2             | 89    | 10        | 10              | 5
Write    | /home/dbadmin | MB/s                 | 412           | 396                             | 206                 | 198                        | 2             | 89    | 9         | 15              | 0
ReWrite  | /home/dbadmin | (MB-read+MB-write)/s | 150+150       | 150+150                         | 75+75               | 75+75                      | 2             | 58    | 40        | 10              | 5
ReWrite  | /home/dbadmin | (MB-read+MB-write)/s | 158+158       | 172+172                         | 79+79              | 86+86                      | 2             | 64    | 33        | 15              | 0
Read     | /home/dbadmin | MB/s                 | 194           | 194                             | 97                 | 97                         | 2             | 69    | 26        | 10              | 5
Read     | /home/dbadmin | MB/s                 | 192           | 190                             | 96                 | 95                         | 2             | 71    | 27        | 15              | 0
SkipRead | /home/dbadmin | seeks/s              | 659           | 659                             | 329.5              | 329.5                      | 2             | 2     | 85        | 10              | 5
SkipRead | /home/dbadmin | seeks/s              | 677           | 714                             | 338.5              | 357                        | 2             | 2     | 59        | 15              | 0

Note: When evaluating performance for minimum and recommended I/O, include the Write and Read values in your evaluation. ReWrite and SkipRead values are not relevant to determining minimum and recommended I/O.