HD Benchmarking (Sandra Buffered Read/Write)

[Fred H]

Hi

I'm currently doing some benchmarking on a hard drive,
and I'm using three different programs:

- Sandra
- Fresh Diagnose
- HDTach

The results I'm getting looks like this (MB/s):

Fresh Diagnose 6.40 Read: 31,3
Fresh Diagnose 6.40 Write: 42,7
HDTach Read: 48,6
HDTach Write: 25,8
Sandra Buffered Read: 83
Sandra Sequential Read: 52
Sandra Random Read: 9
Sandra Buffered Write: 71
Sandra Sequential Write: 47
Sandra Random Write: 24

And now to my question: What is the "buffered read"
and "buffered write" that Sandra performs, that none
of the other programs are testing?

Does any of you have any ide how the different programs
actually test? For me, that information would be almost
as valuable as the results themselves.

Any inputs as to which of the above tests that is the
closes match to "normal office computer usage" would
also be greatly appreaciated.

Cheers
-Fred

 

[Eric Gisin]

I'm currently doing some benchmarking on a hard drive,
and I'm using three different programs:

- Sandra
- Fresh Diagnose
- HDTach

The results I'm getting looks like this (MB/s):

Fresh Diagnose 6.40 Read: 31,3
Fresh Diagnose 6.40 Write: 42,7
HDTach Read: 48,6
HDTach Write: 25,8
Sandra Buffered Read: 83
Sandra Sequential Read: 52
Sandra Random Read: 9
Sandra Buffered Write: 71
Sandra Sequential Write: 47
Sandra Random Write: 24

And now to my question: What is the "buffered read"
and "buffered write" that Sandra performs, that none
of the other programs are testing?

This is why Sandra is not taken seriously. If you open a file
normally/unbuffered, I/O goes though Windows cache. This does not properly
test disk I/O, but does reflect how most applications work.

Does any of you have any ide how the different programs
actually test? For me, that information would be almost
as valuable as the results themselves.

Programs that report STR and access time open the unbuffered
\\.\physicaldriveN device. This gives repeatable, reliable results, but only
for reading.

Others create a test file, usually unbuffered. They time writing and reading
it. The results depend on file placement, whether they preallocate before
writing, other file flags.

Any inputs as to which of the above tests that is the
closes match to "normal office computer usage" would
also be greatly appreaciated.

Not really possible with these benchmarks.