- Logging at Wide Band
- Data Reduction at Fermilab
- Analysis on Workstations
- Backups - System and personal
During the E687 data run we made good use of EXB-8200 2 Gbyte tape drives,
using selected but standard Video grade tape media.
Those drives are obsolete.
We must select a replacement for use in the upcoming E831 data run.
The MOU specifies EXB-8505's, but we should consider newer products,
in light of the report of Fermilab's Serial Media Working Group.
We require storage density at least as good as Exabyte tapes,
substantially better reliability than the EXB-8200 drives,
and reasonable overall operating costs ( drives, maintenance, media ).
We plan to stream data to the drives, clean them on schedule,
and use Vendor approved data-grade tapes, in order to obtain the
Vendor-rated head and drive reliability.
Streaming requires that we log data to disk,
then copy to tape a full tape at a time.
The viable candiate drive types are
Exabyte EXB-8505 (5 Gb)
and
Quantum DLT 2000 (10 GB) .
Drive requirements - 12 EXB-8505 or 9 DLT 2000
- DA831 host --------- 2 EXB or 2 DLT for backups, users
- DB831 calibration--- 1 EXB or 1 DLT for logging/backups
- DD831 Data Logger -- 6 EXB or 4 DLT for logging, spares ( 1.5 MB/sec )
- DE831 Express Farm - 2 EXB or 2 DLT for logging, backups
- possibly several more drives for beam-off usage
- FNAL26 VaxCluster -- 1 EXB for backups, migration
- Misc --------------- half dozen drives on non-FNAL workstations
Tape requirements - 33K EXB or 17K DLT
- DA831 --- about 500 backup tapes
- DB831 --- about 100 Calibration data tapes
- Logger -- about 6,000 EXB or 3,000 DLT
- based on 10 tapes/EXB-drive/day, 3 drives, 200 full days
- alternately, 3 drives * 4000 hours run / (2 hours/tape)
- Express - about 1000 EXB or 500 DLT
- FNUCD --- small
- univ ---- about 1000 miscellaneous tapes
- Analysis 24K EXB or 12K DLT
- 12K out + 1K MOM + 1K Skim + 10K copies
Some other new drives are worth mentioning,
but too new or expensive to consider for immediate E831 usage.
- DLT 4000 20 GB drives started shipping in early 1995.
Data rates are 1.5 MB/s versus DLT 2000's 1.25, so we would need
as many drives as with DLT 2000.
DLT 4000 drives cost $5300, versus DLT 2000's $2800.
Compactape IV media for DLT 4000 costs $100 versus Compactape III at $34.
Drive and media costs seem to be prohibitive.
There is no time for adequate field tests before our run.
- Exabyte Mammoths, at 20 Gb capacity, 3 MB/sec and about $4K drive cost,
could be cost effective.
They may ship in August 1995.
The drives have clear advantages over EXB-8505's,
There is no time for adequate field tests before our run.
- IBM's Magstar 3590 10 GB drives IBM will ship in July.
They write at 9 MB/sec, but cost over $40K .
Vendor's specifications for the two systems.
- Head Life
- EXB_ 16,000 __ DLT 10,000 hours of head/tape motion
- EXB____ 1.8 __ DLT __ 1.2 years of head/tape motion
- Drive MTBF (10% duty cycle)
- EXB 160,000 __ DLT 80,000
- Data transfer rate
- EXB 0.5 MB/s _ DLT 1.25 MB/s
- Drive cost (Fermilab)
- Media cost
- EXB $ 6 ______ DLT $ 34 __Per Tape
- EXB $ 1.20 ___ DLT $ 3.40 Per GB
We discuss here the
costs , and
system choice .
Current Fermilab purchase prices are used.
DLT tape prices are lower than DEC's catalog price of $44 for quantity 1000.
EXB tape prices reflect quantity 200 purchases last year.
8mm Data grade tape costs have been dropping steadily over the years,
from $35 down to $6,
benefitting from the increasing consumer market for 8mm technology.
There seems to be no similar reduction in Compactape III prices.
DLT costs $ 525 K
- drives____ 9 x $2800 = $ 25 K
- media_____ 17K x $33 = $500 K
- maintenance under warranty
EXB costs $ 220 K
- drives___ 12 x $1800 = $ 22 K
- media______ 33K x $6 = $198 K
- maintenance under warranty
- EXB drives and media seem to be the clear winner,
assuming that we use Data grade tapes for improved reliability.
Trying to save another $100K by using Video grade would not void the
drive warranties, but would result in significantly more operational problems.
- DLT seems to cost an extra $300K, which could better go into the Express CPU,
disks, tape, or the End Of Run Party.