example timings

[Wei-keng Liao]

Michael,

This means that the Lustre file system on Horace has only 2 I/O servers 
available (or officially name object storage targets, OSTs). You can check 
with admin for the detailed info. The output of lfs command looks strange 
to me. Here is one of my output:

  % lfs getstripe Lustre/FS_16_1M/ |grep stripe
  default stripe_count: 16 stripe_size: 1048576 stripe_offset: 30

Without the "grep", the command will show all available OSTs.

Usually, parallel I/O performance needs a parallel file system support to 
see the scalability. Otherwise, in your case, multiple compute nodes in 
the application will compete the communication with the only 2 servers.


Wei-keng


On Wed, 18 Jul 2007, michael bane wrote:

> On Wed, 2007-07-18 at 15:19 -0500, Wei-keng Liao wrote:
>> Can you run the following command on the directory where your
>> output files will be stored?
>>
>>    lfs getstripe <dir path> |grep stripe
>>
>> This will show the stripe info of a Lustre directory.
>>
>> Wei-keng
>>
>
> Wei-keng, thanks for helping on this (I'm new to Lustre). Here's the
> output:
>
> ~$ lfs getstripe ~/tmp/test_par_netcdf/ | grep -i stripe
>
> ~$
>
> and if I choose an arbitrary file:
>
> ~$ lfs getstripe tmp/test_par_netcdf/newFile_gather_f90.ncf
> OBDS:
> 0: ost_horace2.nec.1_UUID ACTIVE
> 1: ost_horace2.nec.0_UUID ACTIVE
> tmp/test_par_netcdf/newFile_gather_f90.ncf
>        obdidx           objid          objid            group
>             1         2840580       0x2b5804                0
>             0         2840579       0x2b5803                0
>
>
> all of which means nothing to me I have to admit!
>
> Thanks,
> Michael
>