90 likes | 155 Views
BE-IT Needs. ICE lxbatch Group List of BE-IT Needs. ICE lxbatch Group. More complicate accounting! NAP up to 5 times through fairshare 800 KHS06 (KHepSpec06) measures average throughput - stable unless we get an upgrade
E N D
BE-IT Needs • ICE lxbatch Group • List of BE-IT Needs
ICE lxbatch Group • More complicate accounting! NAP up to 5 times through fairshare • 800 KHS06 (KHepSpec06) measures average throughput - stable unless we get an upgrade • But number of jobs shrinks continuously since the speed of the processors increases • Presently: 66 – 132 jobs in parallel • Next 6 weeks upgrade factor 5 330 – 660 jobs in parallel • Queues: • The shorter the queue the higher the priority • 200 max per user in 1nw & 2nw • 1500 in 2nw und 5000 in 1nw total in parallel • In essence plan generously in advance and discuss with Bernd Panzer • Any adverse experience let Bernd know! • More general problems present in the BE-IT forum.
List of BE-IT Needs • Tools on lxplus (SLC5) excellent support by Steve Traylen! • All relevant Fortran compilers available on Lxplus/Windows/Mac • New NAGLIB installed including special SMP routines • Missing static libraries provided by Harry Renshall • In general no service expertise left in IT! • Harry Renshall & Eric McIntosh honorary members of our group! • All trouble, demands, requests etc BE-IT forum!
Status of Fortran/NAGLIB Intel IFORTmaintained by Andrzej Nowak at CERN in openlab collaboration with Intel. Always the newest versions are made available. There is a script needed to set up usage on LINUX. On Windows/MAC there are packages to be downloaded and installed. I have tested all 3 versions successfully. Only disadvantage is that it will only work at CERN. Laptop owners might get some license loan(?) to be able to use it elsewhere. Andrzej is also willing to help with debugging these compilers via openlab. Presently, there is a conflict of IFORT with the C++library libstdc++.a now solved and provided in the next update. ==> IFORT is ideally maintained for BE! Harry has upgraded the 32bit bit LF95 compiler to the (last) E version. Harry has also provided a node locked 64bit version of LF95. I tested both successfully. Pierre has provided the 32bit version of LF95 under Windows on a special PC. After overcoming my problems with “Remote desktop connection” from my Windows Laptop I could successfully use LF95 for compiling and linking SixTrack.==> BE can now use LF95 again on LINUX and Windows! The NAG f95 (64bit with 32bit capability) is now available directly on LXPLUS (can not be used on local PCs). ==> BE can now use F95 again on lxplus! The NAGLIB version MARK22(21) has been fully installed by Pierre. Things are bit more complicated than for MARK18 since there is a flavor for each relevant Fortran compiler (LF95 is still in MARK21). I have tested all versions successfully in compiling and linking of SixTrack. The NAG experts have pointed out to me that there are special SMP packages that work in conjunction with the NAGLIB. We are authorized to install them at CERN. Harry has promised to install these packages in the near future. ==> NAGLIB/SMP should allow BE to dramatically speed-up the beat-beating tool for the LHC!
Fortran/NAGLIB Maintenance Intel IFORT perfectly maintained – no action needed. A) lf95 32Bit version E on LINUX/Windows is the final version – no action needed. B) lf95 64Bit is pretty new therefore upgrades are to be expected. NAG F95 64Bit is constantly upgraded. NAGLIB including SMP will also be upgraded occasionally. ==> I propose that we review the upgrades once a year and decide what should be upgraded. Occasionally, it might be necessary to upgrade if a serious problem has been encountered. The present trouble with IFORT may serve as an example.
Static Libraries & SVN From IT I have received a recipe for producing the needed libraries to produce a statically linked executable. However, the procedure is very cumbersome and it takes me hours to produce them whenever the packages change. My request is if somebody in IT could provide me with a script to produce these libraries in an automatic way starting from a source SRPM. This could be extremely useful and solve the issue once and for all without having to “piss against the wind”! Due to an apparent misunderstanding with the SVN experts I had to wait for almost 2 months to get an answer concerning a feature of SVN that I thought was a malfunction. Contrary to CVS, subversion does not give default access to the SVN repository for certain commands. At last, after intervention from Nils, I got an answer which allowed me to find the answer via an Internet search: svn status –u tkdiff –r NNNN I hope the SVN service could become more userfriendly!