[ale] Confusing RAID Performance

[Brian Pitts]

I recently finished running iozone benchmarks comparing hardware and
software RAID6 using 11 SATA disks on a server has 8GB of RAM.
There was one part of the results that jumped out at me as not making
sense. For the fread and re-fread reports, the perfomance of the
hardware RAID increased when the test file went from 8GB to 16GB.

Does anyone have a theory about why this might be?

Here is the puzzling data. The top row is records sizes, the left column
is file sizes. See how performance drops dramatically as we go from 4GB
to 8GB, since the test file no longer fits in RAM, but then increases
some for 16GB.

Fread Report	64	128	256	512	1024	2048	4096	8192	16384
4194304	3499493	3559324	3485319	3338358	3353441	2864384	1614729	1537980
1469821
8388608	220831	216510	232010	325999	327661	332615	356501	357100	321870
16777216	644804	644335	629204	506989	624378	595963	596668	606618	652013

Re-fread Report	64	128	256	512	1024	2048	4096	8192	16384
4194304	3501556	3563106	3509425	3339554	3364700	3123160	1611168	1535767
1533804
8388608	234456	226453	242365	381102	386371	407422	389960	442331	384986
16777216	688988	688057	686493	686028	683425	687335	687437	686031	689457

The definition of the fread test is "this test measures the performance
of reading a file using the library function fread(). This
is a library routine that performs buffered & blocked read operations.
The buffer is within the user’s address space."

I have more of the iozone output in this publicly-viewable spreadsheet
if anyone cares to take a look at the rest of the results.

https://spreadsheets.google.com/a/apidb.org/ccc?key=0AoW4-KM82tL9dDV3a3NMUmZ0aWozaFh6RWFRUGdaakE&hl=en&ndplr=1#gid=0

-- 
All the best,
Brian Pitts