Ext3 users - Jan 2007 - [Q] How can the directory location to dd output affect performance?

I have two Opteron-based Tyan systems being supported by PCI-e Areca 
cards. There is definitely an issue going on in the two systems that 
is causing significantly degraded performance of these cards. It 
appeared, initially, that the SATA backplane on the Tyan chassis was 
wholly to blame.

But then I made an odd discovery.

I'm running from the Ubuntu LiveCD for 64-bit. It uses kernel 
2.6.19-7 and the RAID drives are formatted as ext3. The benchmark 
command is dd if=/dev/zero of=output oflag=sync bs=100M count=1

My root is organized has a /maurice directory and a /maurice/drbd 
directory and initially I had changed to that directory to run the 
benchmark. In here, the speeds were slow, averaging about 40 
MB/second.
When I happened to run it from /, I suddenly began getting about 70 
MB/second. So in some bizarre fashion, the location to where the 
output of dd is directed to dramatically impacts the performance. I 
have run from other directories and the performance varies depending 
on which directory I'm in.

Can anyone explain this?
-- 

Maurice Volaski, mvolaski at aecom.yu.edu
Computing Support, Rose F. Kennedy Center
Albert Einstein College of Medicine of Yeshiva University

On Fri, 12 Jan 2007, Maurice Volaski wrote:
> the RAID drives are formatted as ext3. The benchmark command is dd 
> if=/dev/zero of=output oflag=sync bs=100M count=1
------------------^
> My root is organized has a /maurice directory and a /maurice/drbd directory
> and initially I had changed to that directory to run the benchmark. In
here,
> the speeds were slow, averaging about 40 MB/second.
> When I happened to run it from /, I suddenly began getting about 70 
> MB/second. So in some bizarre fashion, the location to where the output of
dd
> is directed to dramatically impacts the performance. I have run from other
> directories and the performance varies depending on which directory I'm
in.
Strange indeed. Only thing that comes to mind is: you're specifying the 
output file not as an absolute path, but relative: the directories (and 
its contents) are distributed all over the disk: some may "live" in
the
inner part of the plattern, some in the outer part - and different areas 
have different speeds. I've never encountered this and I could be dead 
wrong, but I'd suggest to specify the same 'of=/path/to/output' - I 
could imagine that it's more likely that for the next benchmark the 
filesystem uses the same on-disk location...no?

Christian.
-- 
BOFH excuse #12:

dry joints on cable plug