Performance Testing - Please Help

[ElliotA]

I'd like to gather some statistics on Drobo performance under linux
using ext3. Of particular interest to me is copy performance on
DroboPro, but I supposed others might be interested in other
platforms. Could others post their experiences below?

I propose the following tests:

1. Create a single large-ish file and copy it 5 times.
2. Create a directory of many small files and copy them 5 times.

The following command will create a 792MB file named "large" in the
current directory.
dd if=/dev/zero of=large bs=1000 count=830000

Copy the file with:
time cp -a large large_cp

Use the following bash script to create a directory of small files
with a total file size of 790MB.

#!/bin/bash
mkdir small
for i in {0..200}
do
mkdir small/$i
for j in {0..1000}
do
dd if=/dev/zero of=small/$i/$j bs=10 count=50
done
done

Copy the directory with
time cp -a small small_cp

Notes:
I've found that even the dd command is inconsistent - 10MB/s - 17MB/
s.
I believe there is a cache of roughly 512MB in the drobopro which
probably speeds along small copies, which is why I went slightly
larger.
I've also experienced varied performance as the drobo fills. Please
include the result of drobom status, and if possible test once below
50%, once above, once below 1TB full and once above.

[Sloan]

We used iozone to do some performance testing of our drobopro with a
2TB Lun and ext3. You can see the results here:
http://www.warren-wilson.edu/~sloan.poe/iozone-drobo/

Sloan

[ElliotA]

Thanks, I'm still running the small file copy tests, but perhaps a
more standardized test would be more useful. I'm not familiar with
iozone, but it looks like performance is relatively consistent across
your results. How full was your drobo when you ran the tests, and
have you seen performance change as your drobo filled?
-Elliot

[Sloan]

At that point it was not terribly full. Probably a 100-200 G. We
have gotten it up to 480 (26% of the actual drive space). I'll try to
re-run the tests after I've gotten more than 1T on it.

Sloan

[ElliotA]

Results:
drobom status
/dev/sdc:/dev/sdd:/dev/sde /drobo:/drobo2 DroboPro 30% full - ([], 0)

df -h
/dev/sdc1 2.0T 455G 1.6T 23% /drobo
/dev/sdd1 2.0T 748G 1.3T 37% /drobo2

Drobo currently has 1.2TB of data
-----------------------------------------------------------------------
Large file creation:

dd if=/dev/zero of=large bs=1000 count=830000

830000000 bytes (830 MB) copied, 55.8937 s, 14.8 MB/s
-----------------------------------------------------------------------
Small file creation:
time ./create_small_files.sh
real 11m59.466s
user 3m42.720s
sys 4m56.910s
-----------------------------------------------------------------------
Copy large file:

time cp -a large large_cp

real 1m15.297s
user 0m0.030s
sys 0m3.780s

time cp -a large large_cp1
real 1m3.358s
user 0m0.020s
sys 0m3.280s

time cp -a large large_cp2
real 0m58.628s
user 0m0.020s
sys 0m3.640s

time cp -a large large_cp3
real 0m55.309s
user 0m0.060s
sys 0m3.720s


time cp -a large large_cp4
real 0m45.011s
user 0m0.020s
sys 0m4.070s

time cp -a large large_cp5
real 0m47.632s
user 0m0.050s
sys 0m3.590s
-----------------------------------------------------------------------

time cp -a small small_cp

real 9m3.848s
user 0m1.030s
sys 0m17.240s

time cp -a small small_cp1
real 7m57.326s
user 0m1.060s
sys 0m15.480s

time cp -a small small_cp2
real 5m31.517s
user 0m0.930s
sys 0m13.900s

time cp -a small small_cp3
real 8m55.733s
user 0m0.950s
sys 0m15.060s

time cp -a small small_cp4
real 11m49.761s
user 0m1.270s
sys 0m16.390s

time cp -a small small_cp5
real 11m57.760s
user 0m1.140s
sys 0m17.260s
-----------------------------------------------------------------------
Summary:
Copy times seem somewhat more consistent for larger files, but
variation of about 40% still seems pretty inconsistent.