[PVFS-users] Re: Parallel

[Robert Latham]

On Tue, Mar 16, 2004 at 04:11:05PM +0800, Liu Ming wrote:
> Hi,
> 
> I've installed PVFS 1.6.2 on my cluster. When I was trying to test
> this system using PVFS API, pvfs_open, pvfs_write, pvfs_read, I've
> found those I/O nodes are accessed in order, one by one, not
> concurrently. Is there something wrong or I miss something?

I imagine you had a simple program that ran on one machine and
accessed a pvfs1 file.    One program running on one machine will
access a pvfs1 file in chunks, as you are seeing.  

> I'm also trying to set a twin-cluster. It means 2 same clusters are
> mounted on the clients and the data stored on and written parallel
> to  those 2 clusters is the same. Those 2 clusters are the backup of
> the another one. Do you think it is possible that we just use the
> PVFS API and some special settings to do this? Or MPI API is
> necessary on the client nodes?

I'm afraid i don't perfectly understand what you're trying to do. It
sounds like you have two pvfs1 volumes (volume A and volume B), and
your applications are writing their data twice: once to A and then
writing the same data to B.  

If that's what you are doing, there is no better way in software to
achieve redundancy.  

Some groups have had good success with hardware techniques ( raid
volumes on the storage nodes, high availablity configurations).  We
don't have the process well documented right now, but hope to in the
not-too-distant future.

==rob

-- 
Rob Latham
Mathematics and Computer Science Division    A215 0178 EA2D B059 8CDF
Argonne National Labs, IL USA                B29D F333 664A 4280 315B