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

Re: 19990716: CONDUIT latencies

On Fri, 16 Jul 1999, David J. Knight wrote:

> 
> Chiz,
> > 
> > There are some things we can do to tune the LDM requests,
> > and we can look at other upstreams. 
> > 
> > There is currently 1 bottle neck that drives the initial latencies up.
> > This is the T-1 line from NCEP to GSFC which is roughly capable of
> > delivering 550MB per hour. Since the AVN (now out to 84 hours) can exceed
> > 600MB, and the MRF is also pushing 500MB then things can get backlogged-
> > especially when NCEP has Cray crashes and then come back up and post
> > the data all at once.
> 
> Not much we can do about that one (at least for now).


Yes, looking at the latency plots, it frequently goes over 3600s
coming in to Unidata.  So, it follows that its going to be
late everywhere else.

> > 
> > For the past month we have also been downgraded from out 100baseT
> > connection to a 10baseT connection while our building is rewired.
> > When our switch gets moved to the new wiring closet it could improve
> > things as well.

Trying to shove this much data (incoming, plus several outgoing copies)
through the one 10baseT connection to flip.unidata.ucar.edu is probably
also not going to help. 


Considering that >3600 latencies are going to be common for
a while, can LDM be made to hold on to it longer?  An 1hr 10minutes
late is better than never at all...especially when its right near
the edge and drops some but not all a set of products.




--------------------------------------------------------
 David Wojtowicz, Research Programmer/Systems Manager
 Department of Atmospheric Sciences Computer Services
 University of Illinois at Urbana-Champaign
 email: address@hidden  phone: (217)333-8390
--------------------------------------------------------