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

[IDD #DDH-304651]: latentcy problems between idd.unidata.ucar.edu and freshair.atmos.washington.edu

Hi Harry,

re: why WMO feed on single rpc.ldmd request would get shortchanged

> The only thing I can think of is that somehow it is trying to do a per packet
> round-robin on the connections.  Since WMO has more products (and thus more
> packets) than the other feeds, it gets starved.  Since CONDUIT and NEXRAD2 are
> spilt over multiple connections (I have 5 for CONDUIT) the number of products
> per connection is lower that if it was one connection.

> Does this sound like a reasonable hypothesis?

Actually, yes.  Your switching to a kernel has dropped the latencies for
not only the WMO feed, but correspondingly for the NNEXRAD feed, and
NNEXRAD also has about 20K products per hour.

Your findings regarding the introduction of 'congestion control' in Linux
kernels 2.6.13 will be extremely valuable to us as we were just about to
embark on an upgrade of the data server backends of the cluster
idd.unidata.ucar.edu (which are currently running the 2.6.11-x kernel).
We may have to more carefully plan the OS upgrade on those machines since
they are all handling in excess of 100K products per hour.  I will get together
with the IDD group (Mike, Steve, Chiz, and Jeff) about this early next

Thanks for the superb sleuthing!


Unidata User Support                                    UCAR Unidata Program
(303) 497-8642                                                 P.O. Box 3000
address@hidden                                   Boulder, CO 80307
Unidata HomePage                       http://www.unidata.ucar.edu

Ticket Details
Ticket ID: DDH-304651
Department: Support IDD
Priority: Normal
Status: Closed