In the 1990/1992 E687 data run,
the ability to analyze small samples of data quickly in the counting
room played a major role in our ability to keep the quality of the data high.
We also derived some advantage from the "expressline" Silicon Graphics system
in the Feynman Center,
used to run the occasional full data tape to check charm signals, and used to
reconstruct some special calibration runs shortly after they were taken.
For the 1996 E831 data run we are purchasing a dedicated Express Analysis system
for counting room.
The system will have about 600 SPEC92 of CPU power.
The system will be configured with appropriate disk and tape,
for software develpment and data handling.
We plan a data taking rate of about 20,000 events per spill,
at about 5 KBytes per event.
This implies a sustained data rate of about
1.5 MBytes/second and 333 events/second.
The full analysis of an average E687 event requires about 20 SPEC-seconds.
If everything remains the same for E831, a 500 SPEC farm could analyze 7% of the data.
Event sizes may increase due to the new Vertex and Straw detectors, and analysis time
may increase somewhat.
But full calorimeter reconstruction will probably not be available,
nor will it be necessary for most of the Express monitoring.
Streamlined calorimitry will probably bring Express reconstruction time down to half
the full E687 reconstruction time, allowing 15% of the events to be analyzed.
A combination of online and offline information will be used to select
events with charm content, or special monitoring value.
The Express system will write output tapes in DST format, to keep
down the tape and data handling costs.
We will write no more than a tenth as many tapes as the full data stream.
- Fermilab supported system ( AIX, IRIX, Solaris, OSF/1 )
- Both SMP systems and workstation farms are acceptable
- About 48 Megabytes memory plus 16 megabytes per additional SMP processor
- Licensed to support at least 16 interactive sessions
- Software development tools for Fortran and C
- NFS client and server
- At least 500 total
SPEC (sum of SPECs for individual nodes)
- SPEC is SPECnet * (TINYref/SPECref).
- SPECnet is 3/4(SPECint92) + 1/4(SPECfp92),
reflecting the relative usage of Integer and Floating point
in the E831 code.
- SPECint92 and SPECfp92 are the Vendor's published ratings.
When only SPECrate_*92 is available, we divide it by 24 to get
a fair SPEC*92 estimate.
- TINYref means the Specint92-normalized rating given by the Fermilab
TINY benchmark, run on an available Fermilab reference system.
- SPECref is the vendor's published SPECint92 for that same reference system.
- SPEC total is the sum of the SPEC ratings of the individual nodes.
- We use a standard Fortran compiler presently in use at Fermilab
- We use standard IEEE arithmetic (AIX -qrndsngl , Solaris not -fnonstd)
- We use native code generation (IRIX -mips2 , Solaris -cg92 )
- We use No source-level optimization
- We use no user-code inlining
- Reads events from data logger workstation via network at 1.5 megabytes/second
- A single EXPRESS I/O node connects to the Data Logger and serves disks/tapes
- Links to EXPRESS farm nodes may be Ethernet where appropriate.
Necessary network equipment must be included in the bid price.
- Supports 27 Gigabytes of input disk and 18 Gigabytes of output disk.
- Supports Exabyte EXB-8505XL tape drives on dedicated SCSI bus
- Output rate to disk is under 200 kilobytes/second net
Input and output data files will be stored on disk in the Express system.
This decouples Express processing from any tape handling problems.
Each input file will be analyzed by a single process.
A single Express Host system will receive data from the Data Logger
via FDDI or Fast Ethernet.
Input data will be stored on three of the Host's 9 GB disks.
Output from the EXPRESS analysis processed will be written to two 9 GB disks
before spooling to tape.
This is a relatively small system, to be supported with existing
resources.
We require a hardware and software environment already supported
at Fermilab.
We have purchased Digital Alpha based workstations,
based on a competitive bid between Digital, IBM, SGI and SUN.
The selection was made based strictly on the performance/price ratio.
The specific configuration purchased was chosen based on price,
performance, compatibility,
networking options, and supportability.
The base system is listed here.
We are purchasing additional components (disks, tapes, etc).
- host
- Alphastation 5/266, 96 MB, 2 GB disk, 2 MB cache
- 8 plane graphics, 21" Color Monitor
- 2 FWD SCSI
- PCI FDDI
- 16 users, MOTIF, C, Fortran, NFS
- worker node
- Alphastation 5/266, 64 MB, 2 GB disk, 2 MB cache
- 8 plane graphics, 21" Color Monitor
- NFS client
- 4 users