Print

Print


Douglas McNab wrote:
> Hi,
> 
> We have had some issues with file transfer from our WN back to remote 
> SE's. So I have been carrying out some testing to try and replicate the 
> issue.
> I am now seeing strange results from local and remote grid sites with 
> various SE srm implementations:
> 
> The test is a simple simultaneous lcg-cp of a 10M file for varying 
> amounts of WN's .
> 
> Summary:
> lcg-cp's from Glasgow to DPM sites local and remote are 100% successful.
> lcg-cp's from Glasgow to CASTOR/STORM sites are 50% successful.
> lcg-cp's from Glasgow to DCACHE sites are 20-50% successful.

I see below you try 100 connections. The StoRM default setup has a limit 
of 50 simultaneous GridFTP connections.


> 
> Our WN's have lcg_util-1.7.6-1.sl5 installed.
> 
> We initially thought this was network related but now I am not so sure 
> and think it definitely has something to do with the srm it is using.  
> Does anyone know of any current issues with lcg-cp and the varying srm 
> implementations or have you seen anything similar?
> 

I see you tested QMUL's StoRM:

> AS DTEAM TO se03.esc.qmul.ac.uk <http://se03.esc.qmul.ac.uk> (STORM)
> 
> for 98 roughly parallel transfers:
> 
> 49 FAILED - Connection Timed Out
> 49 SUCCESS - Took
> 

So it looks like you hit the 50 simultaneous GridFTP connection limit 
(with presumably one other connection elsewhere).

I actually thought I'd raised this to 100, but clearly wasn't 
successful. Is there a recommended number of connections?

Chris