Table of Contents
Advertisement
2
817.132 MB/s0.000 MB/s817.132 MB/s1994950199495
3
817.160 MB/s0.000 MB/s817.160 MB/s1995010199501
4
817.096 MB/s0.000 MB/s817.096 MB/s1994860199486
HP-AM456-9002A Rev 01
The following table describes the options used in this command. For a complete
list of
utility options, is
perf_test
-s memory-size
-t num-threads
-R
-o
-A async-io-depth
For best results, take the following factors into consideration when running the
utility.
perf_test
Memory Access Patterns
The
memory access modes include random, sequential, and single-
perf_test
address. Accessing just a single address can radically impact the performance
a device can deliver, as can sequential and random access. So, know which
address access mode is being used when running tests.
Threading and Asynchronous I/O
Use
to specify the number of threads to spawn and the number of
perf_test
outstanding I/Os within each request.
The Linux operating system tends to deal with larger asynchronous I/O depths
better than a larger number of threads. High amounts of I/O can be achieved by
understanding the operating system behavior and using a combination of thread
counts and asynchronous I/O.
VMA-series Memory Array Installation and Service Guide
Network-Layer Performance Testing
perf_test
on page 163.
Specifies the system memory size, in GB (if no
units are provided). Recognized unit/size
qualifiers include:
,
,
k
K
m
appropriate binary multipliers.
Specifies the number of threads to invoke. Each
thread is governed by the options specified for
reads, writes, verification, block size, and so on,
and operates independently in terms of
randomness of addresses touched.
Selects addresses to read/write randomly.
Informs
to bypass the buffer cache by
perf_test
using the
flag when opening a
O_DIRECT
connection to the device-path. This option applies
to block devices only and not character devices.
Specifies the number of I/Os to generate using
the operating system's asynchronous I/O API.
,
,
and
for the
M
g
G
175
Advertisement
Table of Contents
