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

20030614: did something change at ULM?



>From: Unidata Support <address@hidden>
>Organization: UCAR/Unidata
>Keywords: 200306111400.h5BE0HLd016841 LDM-6 IDD firewall

Hi Adam and Chance,

I took a look at the real time statistics tornado is reporting back
this morning (Saturday), and I see a substantial change in the
latencies for the HDS feed and the volume of data that was ingested in
that feed (these two things are not entirely separate from one
another).

For reference, take a look at the latency and volume plots for
the HDS feed for tornado at:

http://www.unidata.ucar.edu/staff/chiz/rtstats/siteindex.shtml?tornado.geos.ulm.edu

It appears that something changed at about 7 pm (1 Z) on Friday.  The
HDS latencies which had been consistantly ranging between 1000 and 5000
seconds dropped to zero for about an hour and then climbed back to the
high values until around 7 am (12 Z) when they again went to zero.

At first, I guessed that the volume in the HDS feed may have dropped to
near zero during these periods, but a review of the volume plot for HDS
shows that this was not the case (even though those times did show that
the volume did drop noticably.  In fact, tornado was seen to ingest
_all_ of the products in the HDS feed for the first time since I
installed LDM-6 and tuned up your ~ldm/etc/ldmd.conf files.

Given the times of the latency decreases, I would ordinarily guess
that some other network activity had dropped off substantially
(like downloads of MP3s, for instance), but you/Chance said that
the I2 connection was basically being used only by tornado.

So, the question is if you/Chance did something on tornado or modified
something in the firewall rules that may have caused the drop in
HDS latencies?  If yes, what did you do?

Also, I was going to recommend -- for a test -- that the limiting of
the network bandwidth alloted to tornado be turned off for a day or so
during the week.  If the latencies for all feeds dropped to zero and
remained there for the entire test period, it would pretty much prove
that the latencies were being caused by the bandwidth limiting
mechanism that is in place.

By the way, exactly what is being used to limit the bandwidth to 
tornado (e.g., Packetshaper, etc.)?

Tom