[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: 19990527: Lots of close_wait

On Thu, 27 May 1999, Unidata Support wrote:

> 
> ------- Forwarded Message
> 
> >To: address@hidden
> >From: Randy Weatherly <address@hidden>
> >Subject: Lots of close_wait
> >Organization: .
> >Keywords: 199905271511.JAA07723
> 
> 
>      All,
>      
>      We are using ldm to feed some large data sets (satellite and model) 
>      between two sites across the internet (Western Region HQ and Eastern 
>      Region HQ, NWS).  When doing a netstat -n, we see lots of close_waits 
>      on that connection.  Looking through ldmd.log, we see that that the 
>      ERH machine makes a lot of connections.
>      
>      This problem seems to come and go, probably depending on how well our 
>      connection is over the internet.
>      
>      My question:  is this a problem? 

Hiya,

Not really, the LDM responds to the state of the network. 


 Is it feasible to send large data 
>      files over the internet like this using ldm? 

We have sent large files 100 megabytes in size. The only problem is that
the file must remain entirely in the LDM queue until the transfer is
completed. This can tie up the LDM queue and make the queue requirement
size larger.  The LDM splits large products into 16k sections that are
sent sequential until the complete product is sent.


 What happens to the data 
>      that is being sent if the connection closes?  Does it pick up where it 
>      left off, or does it resend.

The LDM can stand short outages in the connection without loosing it place
in sending a product.  But once the ~5 minute limit is reach, the
connection is broken and the LDM must resend the complete product again.


  ERH seems to be getting all of the files 
>      we are sending.  But I am concerned about the reliability of our 
>      server in this situation.

If the large products are broke into pieces, and a LDM connection breaks,
then it is more effiecent network usage to resend the smaller products
then a large product. If the current configuration is  acceptable, then
leave it.  If degradation starts to occur, maybe splitting the larger
products might be a better solution.

Robb...



>      
>      Thanks for any insight.
>      
>      Randy
> ---------------------------------------------------------
> Randy Weatherly            Unix Sys Admin/AWIPS
> address@hidden  National Weather Service
> 801-524-5120 x284          Western Region Headquarters/SLC UT
> ---------------------------------------------------------
> 
>      
> 
> 
> ------- End of Forwarded Message
> 

===============================================================================
Robb Kambic                                Unidata Program Center
Software Engineer III                      Univ. Corp for Atmospheric Research
address@hidden             WWW: http://www.unidata.ucar.edu/
===============================================================================