[IDD #UQG-528842]: Puzzling IDD latencies from idd.unidata to metfs1.agron.iastate.edu

Hi again Daryl,

Another comment:

I took a look at the real-time stats being reported by 
both metfs1.agron.iastate.edu and metfs2.agron.iastate.edu,
and it seems that the high latencies are being reported for
the highest volume feeds being REQUESTed, CONDUIT and NGRID.
Here is a listing of the volumes being reported by metfs1:

Data Volume Summary for metfs1.agron.iastate.edu

Maximum hourly volume  32333.782 M bytes/hour
Average hourly volume  17422.871 M bytes/hour

Average products per hour     306925 prods/hour

Feed                           Average             Maximum     Products
                     (M byte/hour)            (M byte/hour)   number/hour
NGRID                  9261.417    [ 53.157%]    14084.641    58195.975
CONDUIT                3407.846    [ 19.560%]    12784.612    49240.600
NEXRAD3                2440.436    [ 14.007%]     2914.372   112083.575
HDS                    1204.240    [  6.912%]     1639.485    39299.700
EXP                     525.464    [  3.016%]     1457.611     4047.350
FSL2                    200.574    [  1.151%]      650.261       48.800
FNEXRAD                 117.786    [  0.676%]      139.598      100.650
NIMAGE                   84.579    [  0.485%]      144.219      109.050
UNIWISC                  68.057    [  0.391%]      136.957       43.575
IDS|DDPLUS               65.436    [  0.376%]       79.650    43682.025
NEXRAD2                  46.742    [  0.268%]      114.076       17.250
LIGHTNING                 0.295    [  0.002%]        0.682       56.000

The existence of high latencies for high volume feeds and low latencies
for low volume feeds is a "classic" feature of volume-related bandwidth
limiting on a per-connection basis.

The only ways that we have found to workaround this kind of imposed
bandwidth limiting are:

- work with the campus network group to see if the limiting has been
  imposed locally by mistake and can be altered

- split the feed REQUEST(s) into multiple, disjoint REQUESTs

  This is easily done for CONDUIT as each product has a monotonically
  increasing sequence number, but not so easily done for NGRID.

  The one thing you can do immediately is isolate the REQUEST(s) for
  the highest volume feeds (i.e., don't REQUEST any other feed in the
  same REQUEST).  The next step would be to start splitting the feed
  REQUEST(s) and then run for awhile to see the effect on the reported
  latencies.

I would guess that this latency issue is fairly new.  If this is the
case, I would suspect that the network folks at IASTATE have imposed
some bandwidth controls either intentionally or unintentionally. We
have seen several instances where a campus starts using software from
Palo Alto Networks, and they have unintentionally started limiting
legitimate network traffic.

Cheers,

Tom
--
****************************************************************************
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: UQG-528842
Department: Support IDD
Priority: Normal
Status: Closed
===================
NOTE: All email exchanges with Unidata User Support are recorded in the Unidata 
inquiry tracking system and then made publicly available through the web.  If 
you do not want to have your interactions made available in this way, you must 
let us know in each email you send to us.