[Bioclusters] updated NCBI 2.2.9 RPMs

[Joe Landman]

On Thu, 2004-07-01 at 09:31, Doug Shubert wrote:
> Hello Joe,
> Ran your static 2.9.9.-5 build against a dynamic build from src.
> results are bellow.
> 
> Please elaborate on the static compile flags you used.

Hi Doug:

  You should be able to pull the compilation options from the patches on
the site.  Depending upon the target, We tried to use the best
performing options that resulted in correct execution.  We have some
unit tests in place, and this is why when users have indicated some
problems we are quite interested to get their tests to see if we can
uncover the problems (or if they are problems with our build).

  If you are doing an Opteron run, I believe we are using

	-O3 -msse2 -m64

as our compilation options.  On our machines, the static and dynamic
RPMs give within a few percent of each other in terms of performance on
BBS.  By default, NCBI uses -mpentiumpro on x86 class machines,
including Opteron.  Our patches fix this, and have worked well since
2.2.6.  

What is interesting is that our 64 bit binary is about 20-30% faster
than a similarly compiled 32 bit binary (built on same hardware, with
same compilers, but using the -m32 option and no -msse2).  The moral of
this anecdote (which has some significant amount of analysis behind it,
as will be seen in a few weeks) is that the 64 bit binaries seem to be
better on average, and in number of cases, significantly better.  We are
getting on average about 28% improvement in various BLAST cases, about
30% in HMMer, and about 30-40% in a number of chemistry applications.

You can run the 32 bit versions, but the 64 bit seem to be (with few
exceptions) better performing.

Note:  all the tests were done using GCC.  We are going to try other
compilers as well, and report on them.

All RPMs, including the source are available from
http://downloads.scalableinformatics.com/downloads/ncbi/ .  The patches
should be there as well, though you can always pull down the source RPM,
install it, and run 

	rpmbuild -bp --target=x86_64 ncbi-toolkit.spec

(or ncbi-toolkit-static.spec if you wish to use that).

Joe


> Doug
> 
> SI-Static build 2.2.9-5
> 1,1,1564
> 2,2,1543
> 2,1,1572
> 3,4,3095
> 3,2,3097
> 3,3,3114
> 3,1,3117
> 4,1,796
> 5,1,1554
> 5,2,1558
> 6,1,789
> 
> 2.2.9 src compiled dynamic 
> 1,1,1660
> 2,1,1661
> 2,2,1665
> 3,2,3300
> 3,4,3311
> 3,1,3329
> 3,3,3329
> 4,1,855
> 5,1,1680
> 5,2,1682
> 6,1,851
> 
> 
> Joe Landman wrote:
> > Folks:
> > 
> >   Fixed a number of problems with the spec file and with the build. 
> > Also set up a working static build, so you won't need the same libraries
> > as I have.  Makes the binaries quite a bit larger though.  We are
> > working on getting the source RPM for this and the dynamic build out
> > later on today.  You can pull the latest build from 
> > http://downloads.scalableinformatics.com/downloads/ncbi/ (as usual). 
> > These use the tarball as released from NCBI on 15-June.  
> > 
> > Note that this build corrects the performance problem on Opteron that
> > some people had reported to us.
> > 
> > Joe
> > 
> >   
-- 
Joseph Landman, Ph.D
Scalable Informatics LLC,
email: landman@scalableinformatics.com
web  : http://scalableinformatics.com
phone: +1 734 612 4615