LINKAGE on supercomputers

wijsman at max.u.washington.edu wijsman at max.u.washington.edu
Fri Feb 14 00:43:02 EST 1992


>If anyone is aware of work ongoing or comleted to implement the LINKAGE
>package on a super computer or parallel computer I would be very interested
>to know what is being done, and the status.
>
>Thanks.
>Bob
>
>       ------------------------------------------------------------
>       Bob Cottingham                  Phone:  713/798-4275
>       Cell Biology, M301              Fax:        790-1275
>       Baylor College of Medicine      Email: bwc at bcm.tmc.edu
>       Houston, TX   77030
>       ------------------------------------------------------------
 
We have made some attempt at implementing the LINKAGE package on the San Diego
super computer.  (We were specifically trying LINKMAP).  We got the program
running, but did not get enough of a gain in speed to be worth the trouble of
using the supercomputer instead of our local computers.  The turn around
time when we did timings under the lowest of the 3 priorities was about the
same as we could get on our own computers, although the actual CPU time was
much less.  In order to get turn around time which would make the use of the
supercomputer worth it, we burned up our allocated time very rapidly.
 
So far our diagnoses are the following: 1) the Cray does not have a Pascal
compiler which vectorizes, so we don't gain much by using a Cray (other than
brute force speed); 2) being naive supercomputer users, we may yet have
missed something important which some experienced user might not miss.
 
We are looking into the problem of implementing the algorithms with parallel
processing, but this is in the very early stages.  Parallel processing
might enable us to use idle CPU time on networked machines on campus
without having to go to the supercomputer centers.  We very much need to
find some solution to our very nasty computation problems (late onset
partially penetrant disease with lots of missing individuals, some complex
pedigrees = days of workstation computation for 2-point analyses, and
months for simple multipoints).  We would be delighted to talk to anyone
who has any ideas or experience also beyond our one abortive attempt.
 
Ellen Wijsman
Div. Medical Genetics, RG-25
and Dept. Biostatistics
University of Washington
Seattle, WA   98195
 
wijsman at max.u.washington.edu
or
wijsman at saam (bitnet)  



More information about the Gen-link mailing list