architecture for CDF computing at CNAF

roughly speaking: 1/10th of what is at Fermilab

O(10) farmlets
 One farmlet per data set
   One farmlet = 1 file server + O(10) worker nodes
    file server = dual PC with O(1TB) disk Giga-Ehternet
    worker node = dual PC with 1GB RAM fast-Ethernet

One gateway node
 to handle file submission
 to buffer output data
 dual PC, lot's of RAM (many tasks) Giga-Ethernet
 O(1TB) disk
 Will evolve to "grid gateway", map "globus" job requests
 into specific batch system, collect output from worker nodes
 for pick-up or push

One file server for common areas
 CDF offline software (imported from Fermilab). O(100GB)
 Need to NFS serve to O(100) nodes, although with limited
 requests. Network Appliance ? Maybe shared across all CNAF.
 Needs backup.

One data base server. O(500GB)
 running MSQL version of CDF DB
 somehow syncrhonised with Fnal Oracle server
 Needs backup.


Some interactive login capability:
 O(10) CPU (NIS cluster)
 common NFS mounted home directory
 code/db from same server as farm
 O(10TB) disk, 1/10th as home with backup, rest spool/scratch.

Network to FNAL
 start with O(10Mbit/sec)
 up to O(500Mbit/sec) by 2005-6