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

Re: Comms

Russ,

Thanks for the clarification!  I want to be sure that we include you on
these exchanges.  In fact, two of our folks have been experimenting
with certain aspects of LDM and are going to be chatting with you
briefly about an ITR proposal that we, UCAR and NCSA plan to
submit.  Linda and Ben were in on a conference call this morning
about it.

Kelvin


At 03:40 PM 10/22/01 -0600, Russ Rew wrote:
>Kelvin,
>
>> Quick question:  is all of this stuff done in serial mode, i.e.,
>> does the LDM system allow for overlapping in things like
>> compression, transmission, or is everything done in sequence?  I
>> realize that the data are collected in sequence, but my point is
>> that, if the processing takes longer than the transmission, then a
>> bottleneck exists within the software that might be eliminated by
>> overlapping various functions.  I'm speaking with zero knowledge of
>> how LDM works in this regard, so forgive me if I'm totally out to
>> lunch....
>
>You can certainly be compressing one or more products while
>transmitting others.  The LDM uses a separate process for sending data
>to each downstream LDM, and that process shares the CPU or CPUs with
>all the other processes running on the machine, including a process
>compressing a data product.  So a compression process can be running
>concurrently with a transmission processes.
>
>However, the LDM cannot start sending a partial product before it has
>been completely stored in the in-memory product queue.  So you cannot
>overlap compression and sending the same data product.  The LDM
>protocol requires the size of the product right up front, so you
>cannot send a product before its complete size is known, which would
>be the case if it was still being compressed.
>
>Similarly, an LDM cannot begin relaying a product from an upstream
>site to a downstream site until the product is completely received
>from the upstream site and stored in the product queue.  In this case
>it's not a matter of not knowing its size, but preventing deadlocks
>and other similar problems ...
>
>--Russ
> 

___________________________________________________________________
Dr. Kelvin K. Droegemeier, Professor          E-mail: address@hidden
School of Meteorology                                    Phone:  405-325-0453
Director, Center for Analysis and                   FAX:    405-325-7614
Prediction of Storms                                       CEL:  
405-413-7847                                
University of Oklahoma                                   WWW:  
<http://parker.gcn.ou.edu/~kkd>http://geosciences.ou.edu/~kkd
Sarkeys Energy Center, Rm 1110    
100 East Boyd Street
Norman, OK  73019  USA