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

[IDDBrasil #XPZ-611749]: Re: 20060609: CONDUIT Latency (cont.)

Hi Waldenio,

re: I regret that you were not able to be in Baltimore for the AGU meeting
> I regret this also :-( May be some opportunity arise in September,
> during another WMO meeting in Peru. My intention is to provide training
> courses for others countries of South America in december. So some of
> these peoples can attend here also.

This sounds good!

re: there is a bottleneck somewhere, and we need to work to find it
> I think that the bottleneck lies in Sao Paulo (FAPESP). So we will
> need to "show them" that they are the bottleneck. To do this work,
> I think that FTP can be a tool.

I am not sure that FTP is the best thing to use for this test.  What
we could do is run a separate instance of the LDM on yakov and some
other machine at CPTEC using a port different than 388.  If there is
port-specific packet shaping going on (our theory), then running the
LDM on a different port will demonstrate it conclusively.

Do you (CPTEC) have a spare machine on which we could load and run
the LDM requesting CONDUIT data from yakov.unidata.ucar.edu?  If yes,
we should work together to get a test setup using a different port
and see if the rate limiting exists on the test machine.

In the meantime, however, it will be useful to do the 10-way split
of CONDUIT feed requests AND change the upstream from idd.cise-nsf.gov
to idd.unidata.ucar.edu.  If my theory is correct, the 10-way split
should result in significantly lower latencies.  The reason for this
is that each request will get 10% of the full feed, or about 150 MB/hour.
Since moingobe is able to ingest the NNEXRAD datastream with no latency,
and since the average hourly volume in NNEXRAD is on the order of
150 MB/hour, the ten-way split of CONDUIT should result in 10 connections
carrying about the same load as the single NNEXRAD feed.

Please let me know when you have implemented the 10-way CONDUIT split
and if you can identify a test machine on which we could run LDM tests
using a port other than 388.

> Cheers,
> Waldenio.


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: XPZ-611749
Department: Support IDD Brasil
Priority: Normal
Status: Closed