Disk I/O performance.

[Unix Admin c o R.Porter]

All. I seem to recall a way of generating random data (such as /dev/random_data) that could be used as input for a 'dd' test. (I want to test throughpuot to my array.)

ie. dd if=/dev/random_data of=/home/array/test bs=512 count=1024

Any help will be appreciated.

[RAC_1]

dd if="some_file_of_known_size" of=/home/arry/file bs=1k

Will also do what you want.

[Bill Hassell]

Actually, I don't think the content of the data stream will have any effect at all. Arrays will improve performance by buffering writes and reads but don't look at the data. Similarly, using dd will generate a series of serial I/O operations. The first run with bs=512 and count=1024 will take place fairly quickly but all subsequent runs will be significantly faster as the data will be bufferred by HP-UX. Also, the bs=512 is extremely small and for a lot of data (dozens of megabytes), you will see a significantly faster performance with bs=32k or perhaps bs=128k.

[Unix Admin c o R.Porter]

I should have been more clear with my initial question. I do not want to read my input data from disk. (It would skew my performance results.) Just like you can write to /dev/null to dump output, I want to read input from a source that is not a disk.

[RAC_1 · Accepted solution]

use /dev/null , /dev/zero
/dev/random

[Unix Admin c o R.Porter]

The answer I was looking for --

/dev/zero


Thanks