[Beowulf] NFS over RDMA performance confusion

[Joe Landman]

On 09/13/2012 11:21 AM, holway at th.physik.uni-frankfurt.de wrote:
>
>> I assume so, but just to be clear you witnessed this behavior even with
>> the -I (directio) parameter?
>
> Yes.
>
> for i in 1 2 4 8 16; do /cm/shared/apps/iozone/current/sbin/iozone -I -l
> $i -u $i -r 16k -s 10M -F file1..file16 ; done > output &

-s 10M ???

This is a 10M test file size.

Ok.  Let me be clear, 10M as a file size is not useful as a test.  The 
numbers are, for lack of any better way to describe this, meaningless.

 From the iozone man page

        -s #   Used to specify the size, in Kbytes, of the file  to 
test.  One
               may  also  specify  -s #k  (size  in  Kbytes)  or -s #m 
(size in
               Mbytes) or -s #g (size in Gbytes).


If each storage node has 128GB ram (I think this was the number) and you 
want a "serious" (as in real) test, you need to take your N storage 
nodes @128GB each, sum up the ram, double it, and that is your file 
size.  When you have that, your performance numbers will be uncached, 
and therefore far more likely to represent the actual hardware performance.




-- 
Joseph Landman, Ph.D
Founder and CEO
Scalable Informatics Inc.
email: landman at scalableinformatics.com
web  : http://scalableinformatics.com
        http://scalableinformatics.com/sicluster
phone: +1 734 786 8423 x121
fax  : +1 866 888 3112
cell : +1 734 612 4615