What's new in KP3 computing ? (ARCHIVE 9512)

• December 14^th, 1995
  
  • The disks $12$dka100(DL124) and $12$dka200(DL125), which failed on November 22nd, were reinstalled after the fans had been replaced.
  • The area al_temp0: is now on a 7.0 Gbyte volume set formed by DL124 and DL125. All files were copied.
  • A new area al_temp1: with 3.5 Gbyte was created on DL126. Files on this area have an effective expiration time of 30 days, as compared to one day on other temp areas.
    The new area al_temp1: is mainly meant to hold temporary files of production jobs running several days. Note, that this area is a scratch area, that means no backup, automatic delete. The only difference to al_temp0: is the different time horizon of the cleanup.
  • The KP3$BROOT disk on the VAX cluster went full, the maximum number of files (about 53100) was reached like in May 13^th, 1994. About 1800 files were purged and deleted.

  ---

• November 22^nd, 1995
  
  • The al_temp0: volume set was rearranged:
    • $12$dka100(DL124) (failed September 13th) was reinstalled (with a new 65 Watt power supply).
    • $12$dka200(DL125) was removed because the fan failed during the last week.
    The area al_temp0 resides now on a volume set build from $12$dka100(DL124) and $12$dka300(DL126).
  • After a few hours the fan on $12$dka100(DL124) failed. So this disk was removed, al_temp0: is now on $12$dka300(DL126) alone.

• November 21^st, 1995
  
  • The permanent mailing lists did not work since sometime between November 13th and 17th because they are now routed to clex2 (a cluster alias for axp601 and axp602) instead of axp612. Files and procedures were updated, everything should work again. Note also, that all logical mail addresses pointing to the AXP cluster are now routed to clex2 too. This should however be transparent.

• November 14^th, 1995
  
  • The KY01 files on the MVS Data Disks are now automatically migrated after 4 hours. This should prevent space shortages but, unfortunately, will also interfere with manual recalls.
  • The daily CMS library disribution will now set the /HISTORY and /NOCONCURRENT element attributes of all new elements prior to the distribution using cms_status.com.

• November 8^th, 1995
  
  • The disk $1$DIA540: (accessed as A01_TEMP0:) failed since yesterday evening, developing many bad blocks. The disk was exchanged. Directories were recovered, files were not (it's a scratch disk).

  ---

• October 31^st, 1995
  
  • Created new CMS library MC117.
  • Made disk DL063/109 (4 Gbyte) now available as AL_DATA1.
  • Added search path AL_DATA covering AL_DATA0 and AL_DATA1.
  • New summary of S117 Reduced RAW Datasets.

• October 30^th, 1995
  
  • The local Ethernet got reconfigured and is now connected to a DEChub 900 with PortSwitches. This makes all 6 FDDI to Ethernet bridge port usable and the whole net remotely manageable. The reconfiguration caused, however, a crash of the local part of the VAX cluster (VSBZ and satelites.).

• October 23^th, 1995
  
  • The DBASE, GULIB, SETUP and ULIB reference copy areas were moved to DL175, all REF_* areas are now on this disk.
  • The UCX patch ECO4 was installed to fix the quote problem. The workaround in STASTO_FTP was removed again.

• October 20^th, 1995
  
  • The ASOFT, WTB and WPB reference copy areas were moved to DL175.
  • Made disk DC032 again available as A01_TEMP0, see September 13th and May 20th.

• October 18^th, 1995
  
  • The AXP cluster was rebooted to activate patches for
    • UCX (V3.3:UCXECO2) - This patch should fix the "Kernel stack invalid" crashed we had on the AXP cluster lately. It broke however the FTP client, so STASTO_FTP got fixed with yet another workaround.
    • Fortran - The version T7.0-763 was installed to fix some compiler bugs connected with string handling (see SYS$HELP:DEC_FORT070.RELEASE_NOTES for details).
      Note: The libraries ULIB, GULIB, TFIT, ALIB, S114 and S117 were recompiled with the new compiler, images relinked !
    • Motif (V1.2:AXPMOTF05_U3) - No problems known. Good news is that the DECwindows RESTART is supposed to work again.
  • The new disk array with six QUANTUM XP34300 disks of 8399520 blocks or 4.3 Gbyte capacity each was finally put into use:

       Area      old place   old size      new place       new size
    
      AL_DATA0   DC032        4.30 Gb      DL175/176/177   12.90 Gb
      AL_POOL0   DL125/126    7.14 Gb      DL178/179/180   12.90 Gb
      AL_TEMP0   DL063/109    4.00 Gb      DL125/126        7.14 Gb
    

    Most of the REF_* reference copy areas were moved to the DL175 volume set too, improving link and compile times.

• October 10^th, 1995
  
  • Reenabled usage of the VMEM feature in PAW_S117. The amount of VMEM differs for the _X11 and _XL version and is

       Version    VMEM     virtual size    PFquota left
        _X11       32           94k            158k
        _XL        16          196k             52k
    

    The limitation is the remaining PageFile quota (initially 200k) and not the available address space (500k). Note that the PAW default for VMEM is 10 !
  • Changed Fortran qualifiers used for CMS library compilations (see mms_rebuild) on AXP's to use /check=nooutput. This suppresses the

        %FOR-E-OUTCONERR, output conversion error
    

    error messages. An output conversion error will now produce a field filled with ****, but no error message. ULIB, ALIB and S117 were recompiled.

  ---

• September 19^th, 1995
  
  • The second page file for AXP612 was reactivated (see last point of September 13th) after the free space had dropped below 10% of the primary 800k block file.
  • Increased page file on AXP604 from 600k to 1000k blocks.

• September 18^th, 1995
  
  • Fixed a UCX V3.3 related problem in stasto_ftp. Storing of data should now work again.

• September 15^th, 1995
  
  • The DLT tapes are now stored in the space behind the computers in a little drawer cabinet. It has space for 80 cartridges (or 0.8 Tbyte).

• September 13^th, 1995
  
  • The power supply of disk $12$dka100(DL124), which held the KP3 user disk (kp3$root), broke down at September 12th around 19:00 o'clock. To recover, the AXP's disks were rearranged in the following way:
    • A01_TEMP0: is not longer available. The logical name points now to AL_TEMP0:. Some data has been copied to AL_TEMP0:, most has been discarded.
    • AL_DATA0: was relocated, from $12$dkc0(DL154) to $1$dia540(DC032). Access to the new disk is somewhat slower than to the old one. This might affect PAW link and startup times because this volume holds also all CMS reference copy areas.
    • KP3$ROOT was relocated too, from $12$dka100(DL124) to $12$dkc0(DL154).
    All changes should be transparent, all logicals have been redefined.
  • The new kp3$root disk is larger than the old one (4.3 instead of 3.5 Gbyte) and has therefore a cluster size of 16. Thus files are allocated in chunks of 16 blocks, and even small files occupy at least 16 blocks. This increased the number of allocated blocks for some users by more than 10%. The quota's have been increased somewhat to compensate for this.
  • AXP612 uses now one (instead of two) 800 kblock pagefiles. Load should now distribute between AXP604 and AXP612 and make an increase like on February 27^th, 1995 unnecessary.

• September 8^th, 1995
  
  • The Ethernet interface of the MVS system was fixed. It's throughput had deteriorated since early September to about 50 kbyte/sec. The Ethernet datarate is now back to the normal 300-600 kbyte/sec.
    Note, that VSBZ, AXP601, AXP602 and AXP612 get their data over the CTC connection via rzri6f and were not affected by the Ethernet problems.

• September 7^th, 1995
  
  • A bug in the stage_s***_raw procedures was fixed, staging should now work again on the AXP cluster. The problems were `caused' by an improved DCL syntax checking in VMS V6.2 which converted a wrong, but nevertheless accepted, statement into an error.
  • The node AXP602 is back up, but now with two 5/250 processors and 512 Mbyte memory. These CPU's are currently the fastest in the market and about twice as fast as the AXP604 or a SP2 node.

• September 4^th, 1995
  
  • Our new Alpha, named AXP604 got installed. It has 192 Mbyte memory and should be roughly 1.6 times as fast as AXP612 and even 20% faster as AXP601. It has currently no monitor, therefore it can only be used via Xwindows. AXP604 has currently no local disks and only a 300 Mbyte pagefile.
  • AXP612 had hardware problems to access the 5 disks on the SCSI A adapter. The total cable length was slightly above the limit of 3 m. This together with the new VMS 6.2 SCSI driver supporting tagged comand queuing could have caused the problems of the last days. Therefore the SCSI modification done July 24th was undone, disks DL063,DL109 (AL_TEMP0) were moved from DKA400:,DKA500: back to DKB400:,DKB500:. This solved at least the disk access problems.
  • The disk BZ_TEMP0: got repaired (new power suppy, it broke on August 21st) and was reinstalled on VSBZ.

  ---

• August 30^th, 1995
  
  • The AXP cluster was upgraded to VMS V6.2. This included the following new layered products:
    • DEC Fortran V6.3-711
    • DEC C V5.0
    • DEC TCP/IP Services V3.3
    • File Optimizer DFG V2.1
  • The AXP612 tape drives are now clusterwide accessible under the names AXP612$MKD0: and AXP612$MKD100:.
  • The node AXP601 is now a two processor system with 256 Mbyte. The node AXP602 is currently unavailable.
  • The sources for ULIB, TFIT, ALIB, S114 and S117 were recompiled on the AXP cluster with the new Fortran V6.3.

• August 28^th, 1995
  
  • Switched to CERN Library version V95A on the VAX cluster ! Note, that the AXP cluster was using this version already since June 20th. The The CMS libraries ALIB, S022, S114 and S117 got recompiled and the S022, S114 and S117 PAW images relinked.
    The CERNLOGIN default is now V95A on all VMS systems ! Check out the PAW 2.06 release notes !
    Due to a bug in PAWSHROPT you'll get a linker wanring on VAX systems ("conflicting attributes for psect PAWLUN"). This will be fixed soon.
  • PAWSHR is now an installed image on KP3 Alpha systems.
  • The default extent size of all KP3 disks on the VAX cluster was set to 256 for ROOT disks and 1024 for POOL and TEMP disks. The disks A25_POOL0 and A27_POOL0 got set to 1024 too. See also July 13th.

• August 21^st, 1995
  
  • The power supply of disk BZ_POOL0: (dka300:) died on friday August 16th around 16:00 and caused VSBZ to crash. It was not possible to reboot VSBZ, so the KP3 part of the VAX cluster was unavailable for the whole weekend.
    VSBZ was finally rebooted in Monday morning. The disk power supplies were swapped, making BZ_POOL0: working again but BZ_TEMP0: unavailable.

• August 13^th, 1995
  
  • A power glitch due to a thunderstorm crashed the whole AXP and part of the VAX cluster (VSBZ survived) at 16:40. Most systems came back on August 14th between 10:00 and 11:00.

• August 11^th, 1995
  
  • The CMS library MCLIB was moved to the AXP Cluster.

  ---

• July 24^th, 1995
  
  • Moved disks DL063,DL109 (AL_TEMP0) from DKB200:,DKB300: to DKA400:,DKA500:. They share now the same SCSI port as AL_POOL0 and KP3$ROOT.
  • Connected a BA350 StorageWorks chasis with 6 new QUANTUM XP34300 disks as DKB100,...,DKB600. Capacity per disk: 4.095 Gbyte.

• July 24^th, 1995
  
  • Withdrawn the CE_TEMP0: disk area. The CE_POOL0: shared shared the same disk volume and is now the sole user.
  • Cleanup (means shelving of files) of AW_POOL0, BD_POOL0, BD_POOL1, CE_POOL0 and BZ_POOL0. All [MUDEX] directories have been removed on those file systems.

• July 19^th, 1995
  
  • Enabled access date recording on the AXP KP3$BROOT.

• July 13^th, 1995
  
  • The MVS system switched to TCP/IP version V3R1. Fixed stasto procedures. Staging unreliable....
  • The default extent size on all AXP612 disks was set to a reasonably large value to enhance file creation speed:

            kp3$root       256
            al_prod0      1024
            al_data0      1024
            al_pool0      1024
            al_temp0      1024     (since July 7th)
    	a01_temp0     1024
    

    This means, that files are now allocated in 256 or 1024 block chunks and trancated to the used size when they are closed.

• July 7^th, 1995
  
  • Some system and disk volume parameters were changed:
    • The Virtual I/O cache size was changed back to the default of 3.2 Mb (was 1/16 of physical memory) on AXP612, AXP625 and AXP627.
    • The default disk I/O block count (set rms /block) was set to 96 blocks (was 32). This essentially causes a cache bypass for almost all file accesses, only files smaller than 35 blocks will now be cached.
    • The default network block count (set rms /net) was set to 64 blocks (was 32).
    • The balence set count (balsetcnt) was set to 128 processes on AXP612 (was 226).
    • The default extent size on al_temp0 was set to 1024 blocks.

• July 6^th, 1995
  
  • A new MOSAIC version 2.4A, provided by DEC, was installed. It supports Proxy Gateways (see March 31st) and also a no_proxy filter. The login defaults are setup such that all GSI local traffic goes directly to the server whereas all non-GSI traffic uses the proxy server www-proxy.gsi.de. So local access should behave as before whereas outside (especially overseas) access is often drastically improved.
    The symbols XWWW and MOSAIC have been redefined to point to MOSAIC version 2.4A.
  • The ALADIN quota on al_pool0 and al_temp0 has been lowered to 97% of the disk capacity to prevent excessive fragmentation.
  • VAX and AXP disks have been defragmented. The AXP cluster user root disk (kp3$root) will be defragmented daily, all others when needed.

• July 2^nd, 1995
  
  • A thunderstorm caused a power outage around 17:30. All KP3 nodes went down. All disks have been repaired and read checked. Tape B04010 was being read when the power glitched....