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

[LDM #ZRE-293921]: Dropped products

Harry,

If you look at your rtstats(1) results, Freshair is seeing high latencies in 
the WMO and UNIWISC feeds from 
unidata2.ssec.wisc.edu but not in the CONDUIT feed from Idd.unidata.ucar.edu.  
Tom thinks that this is related
to a previous problem you had:

Tom says "Hi" and asks you to remember the tuning that you did on the kernel on 
Freshair that got rid of the 
high latencies you were seeing on all feeds except CONDUIT a little over a 
month ago.  The reason he makes
this comment is that yor're now again seeing those high latencies.  Wasn't this 
related to some sort of "fairness"
algorithm in the kernel?  It seems to Tom that you're running into the exact 
same thing again.  Plus, it's Friday!  :-)

I don't know about the above problem.  I note, however, that the product-queue 
on Freshair is limited by the amount
of space available for in its data portion (it has more than enough 
data-product slots).  This means that as 
space is needed, the oldest data-products will be deleted to make room for 
more.  Because the data-products from
unidata2.ssec.wisc.edu are arriving so late, I suspect they're being 
preferentially removed to make room for other
data-products.

My suggestion, therefore, is for you to split-up your requests for data from 
unidata2.ssec.wisc.edu the way 
the request for CONDUIT data from idd.unidata.ucar.edu is split-up.   
Apparently, on Freshair, throughput is
greater for more TCP connections than for fewer.

I also suggest that you investigate Tom's contention.

Regards,
Steve Emmerson

Ticket Details
===================
Ticket ID: ZRE-293921
Department: Support LDM
Priority: Normal
Status: Closed