Subject: Tape shift report
Nicola.Zaccheo@lngs.infn.it
Date: Fri Jun 16 2000 - 11:32:32 EDT
This is the my "tape shift" report from 13-JUN up to 16-JUN-2000
During the first part of the shift I worked on the NGC re-processing
optimization. I improved all the scripts...for instance I inserted a PHRASE
run identifier in the TCCHECK job in order to avoid a lot of messages from the
TRK-maker checker.
In addition, we decided to test the possibility to skip the RAW to ZEB
production using the 'SET FILE/ATTRIB=(RFM:FIX,LRL:21600)' option instead.
For this reason I produced a tape with all NGC001 files set in this way
that Erik is going to test using the CALTECH facilities.
I also tested the time required for the full processing of the same NGC003
6 runs, using the same CPU (AXPGS0) with different configurations while
*no other* users under AXPGS0_FAST/BATCH or interactively were taking
significant fracion of CPU.
These are the results (Erik: this should include the answers to your
observations):
Full reprocessing chain (i.e. Robot_copy, copy_gc, drivegc,..)
Runs all copied before launching the DRIVEGC:
allowed runs for DRIVEGC total time for the processing(*)
1) 1 32'39"
2) 3 30'20"
3) 6 32'30"
Runs copied and analyzed one per time. The copy of a new run doesn't
start before the previous run deletion:
4) 1 36'26"
(*) for #1,#2 and #3 this include the time spent to copy the files
from tape to disk
It seems that the best configuration is #2
For the case #4 only (but it should be a good indication) I have the
ranges of the CPU time taking by the 6 main jobs:
Charged CPU time Elapsed time
TC_MON 2'23" - 2'35" 5'17" - 7'51"
RZ_MON 28" - 29" 1'40" - 2'35"
MA_MON 1'24" - 1'30" 3'26" - 5'33"
L1_MON 7" 24" - 38"
L2_MON 2'18" - 2'31" 4'03" - 6'59"
L3_MON 5" 11" - 20"
While I was writing this report, Erik gave me the O.K. for the RAW2ZEB
production skipping. I started NGC001 (DRIVEGC_RD only because we had all the
files on DISK$SCRAUSA7:[MACRODATA.RAW]) and NGC002 (I decided to re-process
completely this last tape).
RAREMAKER:
I launched the RAREMAKER for NRD005/NRD006 that gave no problems (see my
diagnosis for more detailes).
I also launched the RAREMAKER for NRD007/NRD008 that crashed after 359 files
processed (at only 34 files to end!) because the 'source' drive need the
cleaning tape. I wrote an "emergency" script in order to recover the last
34 files. This job is still running (w/o problems at this moment).
I am also working in RARE DSTs production checking differences between
the "real time" and "new" processing. The starting point (run12802) of the
real time production corresponds to NRD011.
DDM production:
This is the Massimo's e-mail about DDM production:
> downloading the last ZEBRA files from the TA90 tapes we found out they had
> a block sizes much higher than the ones reported on the TAPLIS.TOT and
> higher than the ZEBRA file sizes which were downloaded from the hexabyte
> backup tapes as the following run:
>
> ---------------------------------------------------------------------------
> RUN002823.ZEB (from hexabyte backup tape) - Blocks (56658/56665)
> RUN002823.ZEB (from TA90 tape) - Blocks (56827/56840)
>
> TAPLIS.TOT entry: (agree with the hexabyte backup copy)
>
> RUN002823 TAP 56658 21-JAN-1991 05:29 22-JAN-1991 13:09 M0342A
> ---------------------------------------------------------------------------
> After having executed the ZEB2RAW, we got two RAW files with the same block
> sizes without differences but we'll be using the hexabyte backup tapes anyways
> even if we could use both of them (TA90 or hexabyte).
> Furthermore, while we were trying to read the pizzas magnetic tapes, we
> found several TA90 tapes which contain the backup copies of ~70 pizzas and
> these TA90s can be easily read as usual we do.
> For this reason, after having restored all hexabyte tapes, the DDM production
> will continue downloading ZEBRA files from B037N1-N3 to B009N1-N4 TA90 tapes
> which are the last available backup copies.
> The RUN001082.ZEB (17-JAN-1990) will be the last one we'll download from
> TA90 tapes.
In addition to this, we found other TA90 tapes wich contain **ALL** the backup
copies of remaining "PIZZAs" tapes (unfortunatly we didn't find 4 TA90 tapes in
this set). It's important to point out that the PIZZA's drive seems to be not
able to read the tapes...Nazzareno is now working on this.
I have just completed the DDM35 directory that will be copied/backupped during
this night.
This archive was generated by hypermail 2a24 : Fri Jun 16 2000 - 11:32:37 EDT