Version 3.0					September, 1996

fasta30t6 now contains a fully functional version of the pvm
implementation of the fasta programs.  The pvm versions (pvcompfa,
pvcompsw) use exactly the same comparison/alignment functions as the
fasta3(_t) versions.

Most significantly, the program now produces sequence alignments as
well as lists of high scoring sequences.  Thus pvcompfa and pvcompsw
are fully functional replacements for fasta3(_t) which allow large
scale searches of one library against a second library.

Sequence libraries:

Unlike fasta and fastea3(_t), pvcompfa and pvcompsw keep the entire
library sequence in memory (divided among the various machines). This
has two effects: (1) currently, all database sequences must be in a
single file. (2) The maximum number of sequences read by a single
worker node is specified in w_mw.h as MAXSQL.  If your database
contains more than #_nodes * MAXSQL, the remaining sequence will be
ignored.

The query sequence library should be in "fasta" format.  I have done
the most extensive testing with searching libraries in FASTA and PIR
format, although Genbank flat file should work as well.  I have not tested
other library formats with pvcompfa/pvcompsw.

Load balancing:

pvcomplib.c also has some additions that allow the program to be run
efficiently on multiprocessor hosts.  You can now specify the number of
processes you would like to spawn on a single "host" by specifying the
"SP" parameter in the "hostfile" used by pvmd.  Thus, the following lines:

* sp=1004
alpha0.virginia.edu

will cause four worker processes to be started on alpha0.

Likewise, if you have a network with different machines with different
numbers of CPU's, you can use:

alpha1.Virginia.EDU
alpha2.Virginia.EDU
alpha3.Virginia.EDU
alpha4.Virginia.EDU
alpha5.Virginia.EDU
* sp=2204
alpha0.Virginia.EDU

In this example, the "*sp=2204" also indicates that each CPU of the
alpha0 machine is 2X faster than the alpha1-5 machines, so 2X as many
sequences will be loaded on each CPU of alpha0.

pvcomplib.c uses the last two digits of the "sp=" specification to
specify the number of CPU's.

Program limitations:

	At the momemt, only fasta (pvcompfa/c.workfa) and ssearch
(pvcompsw/c.workgsw) are implemented.  Clearly fastx and tfastx need to be
included.

Please send bug reports to: wrp@virginia.edu.

Bill Pearson