[PVFS2-users] More question on unstable pvfs2 on a cluster

Robert Latham robl at mcs.anl.gov
Thu Apr 7 17:46:59 EDT 2005


On Thu, Mar 03, 2005 at 06:43:05PM -0500, Xiuduan Fang wrote:
> Here are more information. PVFS2 is installed on the first 10 nodes on the 
> cluster. These nodes are both I/O servers and clients. The first node is 
> the only metadata server. I noticed that pvfs2-client on the first node 
> consumes more than 100 MBytes memory. This is terrible. 

We get reports of this behavior from time to time.  Unfortunately, we
have not been able to reproduce this runaway memory usage on our own
machines, so it becomes rather hard to track down.  You can restart
pvfs2-client when the memory used becomes too big.  Not a good 
long-term answer I know, but until we find what consumes so much
memory, it's the best we can offer at this time. 

> Also, I am 
> wondering how to know the file distribution in the I/O servers. If I use 
> the following command,
>    pvfs2-cp -t -b 32768 /home/xf4c/testfile/test_1G /pvfs2
> 
> does it mean /pvfs2/test_1G will be distributed on all 10 I/O nodes. I am 
> guessing I can use -n to control how many I/O servers we use but I don't 
> know if we can control the starting I/O server's id. More specifically, is 
> it possible for me to let the file distributed from node 3 to node 7?  
> Thank you for your help.

In CVS, you can pass parameters to pvfs2-cp that specify number of
servers to use (-n) and the strip size (-s), but we don't have a way
to specify the starting server yet.  I'll put it on the list of things
to do, but might not be able to work on it for a little while.

==rob

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


More information about the PVFS2-users mailing list