What's new in KP3 computing ?(ARCHIVE 9409)

---

• September 30th, 1994
  
  • The new DEC Fortran V6.2 compiler installed with VMS 6.1 on the VAX'es does more consistency checks than the old VAX Fortran. One gets

        %FORT-I-VARUNUSED, Variable was declared but not used
        %FORT-I-USEBEFDEF, Use of variable before definition;
        %FORT-I-ASFUNUSED, Statement function was defined but not used
        %FORT-I-CHACONCONTD, Character, Hollerith or RAD50 constant continued across
        %FORT-I-BRNCHINTOBLK, Questionable branch into loop or block 
        %FORT-W-USEUNIVAR, Use of uninitialized variable
    

    The compilers on VAX and AXP are supposed to be very similar in the way they treat those informational errors, but they aren't. The AXP Fortran does currently not issue VARUNUSED messages, but the VAX Fortran does.
    The following modifications were done:
    • VARUNUSED: The mms_rebuild procedure was changed to suppress the generation of VARUNUSED messages. They are truely uncritical.
    • USEBEFDEF: Sources were fixed; added DATA and SAVE.
    • ASFUNUSED: Sources were fixed. Function commented out.
    • CHACONCONTD: Sources were fixed.
    • BRNCHINTOBLK: Very bad style. Should be fixed...
    • USEUNIVAR: Sources were fixed. This is a serious bug !
    All sources of all CMS libraries were recompiled on VAX.

• September 29th, 1994
  
  • New version of DECset on VAX. CMS Reference Copies recreated.
    On September 26th new versions of DECset (LSEDIT, CMS,...) were installed. CMS changed to Version V3.6-10. DEC in it's infinite wisdom choose to change the CMS history header format again, this time from "VAX/DEC CMS" to "DEC/CMS". All history headers now start like

        $! DEC/CMS REPLACEMENT HISTORY, Element CMS_COPY_ELEMENT.COM
    

    All CMS libraries on VAX have been fixed with verify/repair. The history header format is now the same on VAX and AXP.

• September 22th, 1994
  
  • The operating system on the VAX'es was upgraded to VMS V6.1.

• September 19th, 1994
  
  • The default RMS I/O buffers size (set rms /blo=..) is set to 32 blocks on VAX systems. The network buffer size ( set rms/net=..) was reduced from 64 to 32 on VAX and AXP systems.
    This should improve I/O cache efficiency (/blo) and prevent some buffer space bottlenecks (/net).

• September 9th, 1994
  
  • A large number of DCL Scripts use now sys$scratch for temporary files and thus to a temp disk (see August 19th, 1994). Before those files were put into the current directory.

• September 8th, 1994
  
  • Changed error handling in many DCL procedures. Use now on error instead of on severe in all on statements.

• September 7th, 1994
  
  • Leftovers of Mosaic temporary and LSEDIT journal files were deleted from kp3$broot on VAX and AXP, a total of 500 files ! Note, that those files are now directed to bz_temp0: or al_temp0: due to the redefinition of sys$scratch (see August 19th, 1994).

• September 6th, 1994
  
  • Changed some system parameters on AXP systems:
    QUANTUM set to 2 on all our AXP's. (better response ?)
    PFCDEFAULT set to 32 on AXP625. (better I/O caching ?)

  ---

• August 30th, 1994
  
  • The CMS libraries asoft, beam, mclib, s022 and s031 were moved from kp3$aroot to kp3$broot.
  • The problem with batch queues and a CPU time limit (see July 5th) on the AXP cluster has been fixed. The queue SYS$FAST_INF_AXP612 has been removed.

• August 23th, 1994
  
  • Setup central mailing lists for KP3, ALADiN and experiment subgroups.

• August 22th, 1994
  
  • The CMS libraries TFIT, ALIB and S114 have been moved to the AXP cluster.

• August 19th, 1994
  
  • The logical sys$scratch is now set to bz_temp0:[username] on VAX and al_temp0:[username] on AXP.
  • Mosaic uses now sys$scratch for its temporary files.
    (the *tmpDirectory resource is now set to sys$scratch).

• August 18th, 1994
  
  • Activated new HTTP server release. We use now CERN 3.0beta.
  • Fixed IP Domain Name Service setup on vsbz. Now is rzri6a primary name server.

• August 16th, 1994
  
  • The lib_update procedure was rewritten. See revison history.

• August 6th, 1994
  
  • Almost all logical names created by login_aladin are now in the job logical name table. Should improve subprocess creation performance.

  ---

• July 28th, 1994
  
  • The local disks in AXP625 and AXP627 are available under a25_pool0: and a27_pool0: respectively.
    Note: The disks are only mounted locally. The logicals are defined on all nodes, remote access is through DECnet and not very efficient.
  • Re-enabled pagefile on DISK$DL094 for VSBZ because the system was overloaded with 13 users and 84 processes.

• July 27th, 1994
  
  • More Alpha's and Xterminals available:
    • two new KP3 systems: AXP625 and AXP627
    • new central servers: AXP601 and AXP602
    • a new Xterminal XWTEI
    • DSAD has been retired; good bye ULTRIX
    Details in the resource guide

• July 21th, 1994
  
  • AXP cluster down or partially disabled since yesterday.
  • Finalized disk rearrangements:

     Disk    Size    former    New usage
    -----------------------------------------
    DL013    660M  ag_pool1    send for repair; use as Lynx system disk
    DL015    660M  ah_pool1    Used as ce_pool0/ce_temp0 (Volume set DL015/DL061)
    DL061   1.01G  bl_pool0     -- see above --
    DL063   2.00G  bl_pool1    Used as al_data0 (Volume set DL063/DL109)
    DL109   2.00G  ce_pool0     -- see above --
    DL064   1.00G  bl_temp0    renamed to bl_pool0 !
    DL110    415M        --    Now available as ce_pool1		

• July 20th, 1994
  Changed to CERNLIB version 94A on VAX and AXP cluster:
  • The main program is very different, it doesn't link against the version 93D libraries !
  • All libraries (ULIB,ALIB,S114,S022,TFIT,...) were recompiled.
  • With version 94A we have a sharable image on AXP too, so link times as well as image size is drastically improved.
  • The 94A sharable images produce on VAX and AXP a linker warning which can be ignored: %LINK-W-SHRWRNERS, compilation warnings in shareable image file CERN_ROOT:[LIB]PAWSHR.EXE
  • There is are various PAW and a PAW++ executables for S114 and S022 available for VAX and AXP: Exp Executable VAX AXP build with ------------------------------------------------------------------------ S114 PAW_S114_X11.EXE y y @ref_s114:paw_s114_x11.bld S114 PAW_S114L_X11.EXE y - @ref_s114:paw_s114l_x11.bld S114 PAWPP_S114.EXE y y @ref_s114:paw_s114_x11.bld paw++ S114 PAWPP_S114L.EXE y - @ref_s114:paw_s114l_x11.bld paw++ S022 PAW_S022_X11.EXE y - @ref_s022:paw_s022_x11.bld S022 PAW_S022L_X11.EXE y - @ref_s022:paw_s022l_x11.bld

• July 19th, 1994
  
  • Tuned the FTP connection between AXP612 and the MVS system. Peak data rates are now 950 kb/sec. See test report.
  • Deleted many files on the KY01 data account:

     dataset             files    Gbytes   in HSMOPR   Comment
    
     LMDV.S114*.RAW*      1045    19.950       0.000   Reduced data now PDS packed
     DSTV.*                404    17.853      15.164   Also available as DSTX
                          ----    ------      ------
     Total                1449    37.803      15.164  

• July 18th, 1994
  
  • Bug in the PAW command util/rwn/cwn fixed. CWN's with selections should work now on.
  • The Ethernet connection to the MVS system failed in Sunday, July 17th around 5:00am (like last weekend). All Stage requests worked only from VSBZ and AXP612 (with use the MVS16 path) and failed from all other nodes.
  • Internet connections to the U.S. are down since about noon.

• July 15th, 1994
  
  • Tested DEC FORTRAN on AXP
  • Linking PAW_S114_X11 on AXP now completes without the

    	%LINK-W-NUDFSYMS, 1 undefined symbol:
    	%LINK-I-UDFSYM,         KLNKADDR   

    linker error message! All C routines linked with PAW have to be compiled with the qualifiers

    	/prefix=all/extern=common	

    The ulib DCL procedures kuipc_cdf_c_obj and mms_rebuild as well as the s114 DCL procedure rebuild_compile were changed accordingly.

• July 14th, 1994
  
  • VAX Cluster hungup at 8:30. Cluster reboot...
  • Quota's on AL_POOL0 readjusted: ALADIN 5000k -> 6220k blocks
  • Former disk BL_POOL1 temporarily restored under $12$DKA0:[DL063.ALADIN...].
  • All S114 reduced datasets are now packed in PDS for improved access speed. The effictive throughput increases by a factor of 2-3 for 10-20 Mbyte datasets.
  • Tested DEC Linear Tape DLT2000

• July 13th, 1994
  
  • Massive heat problems with our disks (outside 34 degree): the plasic feet of one 2 Gbyte disk on VSBZ were molten ! All 2 Gbyte disks got an improved air circulation (larger holes in the housing). The disk DL126 (AL_TEMP0) had a broken fan.
  • Redirection of queue SYS$FAST_INF_AXP612 to SYS$FAST_AXP612 on AXP612, introduced July 5th, has been removed due to boot problems. Submit explicitely into queue SYS$FAST_INF_AXP612 to avoid the CPU limit problem.
  • AXP601 and the whole AXP cluster was down all day.

• July 12th, 1994
  
  • Using LAT to and from AXP612 works now.
  • Added a small (415 Mb) disk on VSCE. Accessed as CE_POOL1:.
  • Replaced the 2 Gbyte disk holding CE_POOL0 and CE_TEMP0 by a 1.7 Gbyte volume set of two disks. The change should be transparent.

• July 11th, 1994
  
  • The Ethernet connection to the MVS system failed in Sunday, July 10th around 4:00am. All Stage requests worked only from VSBZ and AXP612 (with use the MVS16 path) and failed from all other nodes.

• July 7th, 1994
  
  • The IP traffic between AXP612 and MVS16 is now really using FDDI (IP routing for AXP612 and VSBZ was corrected on MVS).
    First tests show a FTP throughput between AXP612 and MVS16 of 500-800 kbyte/sec for read and up to 950 kbyte/sec for write !
  • New subcommands for MOPER ALADIN:
    RESTART_OS9_AF - Restart OS9 disk server
    UCX_PING - Probe response of an InterNet node

• July 5th, 1994
  
  • The Alpha AXP612 got connected to FDDI. DECnet and IP addresses and names are unchanged, almost all traffic runs on FDDI instead of Ethernet now. FDDI has 10 Mb/sec bandwidth, Ethernet just 1 Mb/sec ! First test results (with a 25 Mbyte dataset on July 5th, 15:30):

       Protocol	from	to	Datarate	CPU Overhead
       LAVC		AXP612  AXP601	1600 kb/sec	about 10% 
       FTP		AXP612  AXP601	 950 kb/sec	about 12%

    The connection to the MVS system has not improved in throughput yet, typical rates are 200-400 kb/sec. Need some tuning...
  • As a temporary fix for the CPU limit problem of batch queues: A new queue SYS$FAST_INF_AXP612 has been setup which has infinite CPU limit and a base priority of 3. The queue SYS$FAST_AXP612 has been redirected to SYS$FAST_INF_AXP612.

• July 4th, 1994
  
  • We had VAX cluster problems over the weekend. VSBZ crashed twice and didn't operate properly on Saturday and Sunday. VSCE, VSAW and VSBD had to be rebooted at 11:40 to get KP3$BROOT out of mount verification.

• July 1st, 1994
  
  • Due to a typo in LOGIN_ALADIN.COM we did not use the fast block multiplex channel connection to the MVS system for stage jobs since 31-MAY-1994 (VAX) and 1-JUN-1994 (AXP).
    The problem has been fixed now. All stage and store activities initiated from VSBZ, AXP601 or AXP612 use now the block multiplex channel connection and connect to MVS16.