进入www.iozone.org的官网，下载安装包，安装包的格式有很多，rpm,tar,gz，bz等
这里讲下tar包的安装，
#cd 版本目录/src/current/
#make linux
这个时候就看到iozone文件生成了，可以直接运行。

./iozone -a -n 4g -g 4g -i 0 -i 1 -i 5 -f /sde/iozone -Rb ./iozone.xls

注意：

1，  写测试文件时，由于文件小，内存足够，不存在数据刷新到硬盘的操作，数据被写到页高速缓存即返回；

2，  读测试文件时，由于数据已在高速缓存，亦可直接返回

 所以当测试文件比机器内存小的时候，读写效率相当高，根本没经过磁盘一级，不能真正反应磁盘性能。提议测试文件的大小一般设置为内存的1到1.5倍。

    Usage: iozone [-s filesize_Kb] [-r record_size_Kb] [-f [path]filename] [-h]
                  [-i test] [-E] [-p] [-a] [-A] [-z] [-Z] [-m] [-M] [-t children]
                  [-l min_number_procs] [-u max_number_procs] [-v] [-R] [-x] [-o]
                  [-d microseconds] [-F path1 path2...] [-V pattern] [-j stride]
                  [-T] [-C] [-B] [-D] [-G] [-I] [-H depth] [-k depth] [-U mount_point]
                  [-S cache_size] [-O] [-L cacheline_size] [-K] [-g maxfilesize_Kb]
                  [-n minfilesize_Kb] [-N] [-Q] [-P start_cpu] [-e] [-c] [-b Excel.xls]
                  [-J milliseconds] [-X write_telemetry_filename] [-w] [-W]
                  [-Y read_telemetry_filename] [-y minrecsize_Kb] [-q maxrecsize_Kb]
                  [-+u] [-+m cluster_filename] [-+d] [-+x multiplier] [-+p # ]
                  [-+r] [-+t] [-+X] [-+Z] [-+w percent dedupable] [-+y percent_interior_dedup]
                  [-+C percent_dedup_within]
 
           -a  Auto mode
           -A  Auto2 mode
           -b Filename  Create Excel worksheet file
           -B  Use mmap() files
           -c  Include close in the timing calculations
           -C  Show bytes transferred by each child in throughput testing
           -d #  Microsecond delay out of barrier
           -D  Use msync(MS_ASYNC) on mmap files
           -e  Include flush (fsync,fflush) in the timing calculations
           -E  Run extension tests
           -f filename  to use
           -F filenames  for each process/thread in throughput test
           -g #  Set maximum file size (in Kbytes) for auto mode (or #m or #g)
           -G  Use msync(MS_SYNC) on mmap files
           -h  help
           -H #  Use POSIX async I/O with # async operations
           -i #  Test to run (0=write/rewrite, 1=read/re-read, 2=random-read/write
                 3=Read-backwards, 4=Re-write-record, 5=stride-read, 6=fwrite/re-fwrite
                 7=fread/Re-fread, 8=random_mix, 9=pwrite/Re-pwrite, 10=pread/Re-pread
                 11=pwritev/Re-pwritev, 12=preadv/Re-preadv)
           -I  Use VxFS VX_DIRECT, O_DIRECT,or O_DIRECTIO for all file operations
           -j #  Set stride of file accesses to (# * record size)
           -J #  milliseconds of compute cycle before each I/O operation
           -k #  Use POSIX async I/O (no bcopy) with # async operations
           -K  Create jitter in the access pattern for readers
           -l #  Lower limit on number of processes to run
           -L #  Set processor cache line size to value (in bytes)
           -m  Use multiple buffers
           -M  Report uname -a output
           -n #  Set minimum file size (in Kbytes) for auto mode (or #m or #g)
           -N  Report results in microseconds per operation
           -o  Writes are synch (O_SYNC)
           -O  Give results in ops/sec.
           -p  Purge on
           -P #  Bind processes/threads to processors, starting with this cpu
           -q #  Set maximum record size (in Kbytes) for auto mode (or #m or #g)
           -Q  Create offset/latency files
           -r #  record size in Kb
              or -r #k .. size in Kb
              or -r #m .. size in Mb
              or -r #g .. size in Gb
           -R  Generate Excel report
           -s #  file size in Kb
              or -s #k .. size in Kb
              or -s #m .. size in Mb
              or -s #g .. size in Gb
           -S #  Set processor cache size to value (in Kbytes)
           -t #  Number of threads or processes to use in throughput test
           -T  Use POSIX pthreads for throughput tests
           -u #  Upper limit on number of processes to run
           -U  Mount point to remount between tests
           -v  version information
           -V #  Verify data pattern write/read
           -w  Do not unlink temporary file
           -W  Lock file when reading or writing
           -x  Turn off stone-walling
           -X filename  Write telemetry file. Contains lines with (offset reclen compute_time) in ascii
           -y #  Set minimum record size (in Kbytes) for auto mode (or #m or #g)
           -Y filename  Read  telemetry file. Contains lines with (offset reclen compute_time) in ascii
           -z  Used in conjunction with -a to test all possible record sizes
           -Z  Enable mixing of mmap I/O and file I/O
           -+E Use existing non-Iozone file for read-only testing
           -+K Sony special. Manual control of test 8.
           -+m  Cluster_filename   Enable Cluster testing
           -+d  File I/O diagnostic mode. (To troubleshoot a broken file I/O subsystem)
           -+u  Enable CPU utilization output (Experimental)
           -+x # Multiplier to use for incrementing file and record sizes
           -+p # Percentage of mix to be reads
           -+r Enable O_RSYNC|O_SYNC for all testing.
           -+t Enable network performance test. Requires -+m
           -+n No retests selected.
           -+k Use constant aggregate data set size.
           -+q Delay in seconds between tests.
           -+l Enable record locking mode.
           -+L Enable record locking mode, with shared file.
           -+B Sequential mixed workload.
           -+D Enable O_DSYNC mode.
           -+A #  Enable madvise. 0 = normal, 1=random, 2=sequential
                                  3=dontneed, 4=willneed
           -+N Do not truncate existing files on sequential writes.
           -+S # Dedup-able data is limited to sharing within each numerically
                 identified file set
           -+V Enable shared file. No locking.
           -+X Enable short circuit mode for filesystem testing ONLY
               ALL Results are NOT valid in this mode.
           -+Z Enable old data set compatibility mode. WARNING.. Published
               hacks may invalidate these results and generate bogus, high
               values for results.
           -+w ## Percent of dedup-able data in buffers.
           -+y ## Percent of dedup-able within & across files in buffers.
           -+C ## Percent of dedup-able within & not across files in buffers.
           -+H Hostname    Hostname of the PIT server.
           -+P Service     Service  of the PIT server.
           -+z Enable latency histogram logging.