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

Re: Data latencies: could much of this be avoided?

Hiya,

A little clarification is needed on the idea of splitting the request to
an upstream host using the hostname and the IP number. This does make two
receiver processes to the upstream host, but it really doesn't make the
LDM run any better. This is an network factor that nobody really
completely understands. It seems that routers on the network try to
perform network balancing according to request, therefore it is assumed
that two request will get more network bandwitdth verses one. Therefore
two LDM processes may get better better reception than one. But, this is
according to the routers configurations that are different all over the
network. So one site splitting the feeds will get better reception and
other sites may not. Again this is all speculation from our system admin
and the network folks.  Also, running two processes may cause a cpu burden
on a machine that is close to a threshold to cause higher data latencies.
The bottom line, try splitting the feeds if the latencies are high but if
one doesn't get better reception, then return to the original
configurations. At least you tried.  Also, pqact is an LDM queue reading
so it's completely out of this loop.

Robb...

On Tue, 14 Dec 1999, Gilbert Sebenste wrote:

> Chris,
> 
> > Ok, I've changed my setup but I'm still getting IDS|DDPLUS|HDS latencies
> > over 15 minutes right now.  Here's what my request looks like:
> >          
> >     51  request IDS|DDPLUS      ".*"    ns2.rutgers.edu
> >     52  request HDS             ".*"    165.230.4.74
> >     53  request MCIDAS          ".*"    yang.sprl.umich.edu
> >     54  request FSL2            ".*"    141.213.23.51
> > 
> > Futhermore, I see only one pqact running:
> >  
> > pluto.ldm> ps -aef | grep pqact
> >      ldm 23340 23337  0 19:54:02 ?       12:28 pqact
> > 
> > What do I need to correct?  
> 
> Nothing. If you re-started the LDM, you're fine. Ldm.pq does all the
> work.
> 

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