[Pvfs2-users] File affinity and CPU resource scheduling concernswhen using compute node as a PVFS server

[Kyle Schochenmaier]

I think we have two different threads here...

As an additional note to what Phil mentioned just now, try :

`dd if=/mnt/pvfs2/abc.ts of=~/ bs=65536`

This will give you an idea of some better performance that is
achievable through the vfs mount for pvfs2.

~Kyle

Kyle Schochenmaier



2009/10/22 Phil Carns <carns at mcs.anl.gov>:
> Hi,
>
> The cp command line tool is likely using much smaller access sizes than
> pvfs2-cp to transfer data.  You can confirm by running cp in strace to
> look at what sizes the read() or write() operations are.  In pvfs2-cp
> you can toggle the access size with the -b option.
>
> -Phil
>
> л³¤²Å wrote:
>> Hi dear Rob:
>> I use two I/O data servers, one of them acts as metadata server,
>> when i run pvfs2-cp /mnt/pvfs2/abc.ts ~/
>> the read speed can reach 11MBytes/s
>> but when i mount the pvfs2 system to /mnt/pvfs2 entry
>> and then run cp abc.ts ~/ the speed only can reach 4Mbytes/s .
>> how can i solve this problem ???
>> i know it's slower ,but why only reach 50 persent .
> _______________________________________________
> Pvfs2-users mailing list
> Pvfs2-users at beowulf-underground.org
> http://www.beowulf-underground.org/mailman/listinfo/pvfs2-users
>