System Configuration Diagram

  Tablespace sizes were increased to support the Injection rate.
  Scripts to create the DB are included in the FDA.
  Table and index initrans were modified.
  The M_INVENTORY, M_WORKORDER, M_PARTS, O_ORDERLINE tables were
  horizontally partitioned.
  Automatic segment space management was used for all tablespaces.
  System managed undo was used.
  Table locks were disabled during the run using disable_lock.sql

  The load program was not modified.

  The following Beans were deployed in READ_COMMITTED mode as per the
  benchmark requirement:
  Mfg      - AssemblyEnt, BomEnt, ComponentEnt, LargeOrderEnt, PartEnt
  Orders   - ItemEnt
  Supplier - SupplierCompEnt, SupplierEnt
  
  All other beans were deployed using pessimistic locking mode such that all
  selects including finders are issued with a FOR UPDATE clause in select
  statement:
  Corp     - CustomerEnt, CustomerInventoryEnt
  Mfg      - InventoryEnt, WorkOrderEnt
  Orders   - OrderEnt, OrderLineEnt
  Supplier - POEnt, POLineEnt, SComponentEnt
  Util     - SequenceEnt
  
  The ItemEnt bean was cached for 20 minute intervals by setting:
  read-only to true, and
  time-to-live to 1200000 (ms)
  in orders-toplink-ejb-jar.xml deployment descriptor.

  To ensure database durability, RAID 1 (mirroring) was used for the
  data files and database Redo logs.
  The application server was using mirrored disks to
  provide durability for server logs and JMS persistence filestore.

  An 80 minute run at an injection rate of 955 increased storage by 2300MB.
  This extrapolates to 44G of storage for a 24 hour run.
  The database is configured with over 1000G of durable storage.

  Oracle Application Server 10g uses pass-by-value as required by the EJB
  specification.

Supplier  Description                                   Product #          Qty
--------  -------------------------------------------   ----------------   ---
HP        HP rx2660 2 dual-core 1.6GHz/18MB processors  AD245A-003           2
HP        rx2660 PCI-X 3 slot cage option               AD246A               2
HP        8GB DDR2 memory pair (2 x 4GB DIMMs)          AD276A               4
HP        36GB 10K RPM drives                           AD140A               6
HP        Dual port 1000BaseT LAN adapter card          A7012A               2
HP        Dual Port 4Gb Fibre Channel Adapter           AB379A               2
HP        HP-UX Integrity FOE w/Sys 2 Proc PCL LTU      B9430AC              8
HP        3 year 24x7 hardware and software support     HA110A3

HP        HP rx6600 4 dual-core 1.6GHz/24MB processors  AD134A-180           1
HP        8GB memory using 2GB DIMMS                    AB565A              12
HP        73GB 10K RPM drives                           AD141A               2
HP        Dual port 1000BaseT LAN adapter card          A7012A               1
HP        Dual Port 4Gb Fibre Channel Adapter           AB379A               2
HP        HP-UX Integrity FOE w/Sys 2 Proc PCL LTU      B9430AC              8
HP	  3 year 24x7 hardware and software support     HA110A3

HP        Modular SAN Array 1000                        201723-B22           4
HP        72GB 15K U320 Pluggable Hard Drive            286778-B22          44
HP        3 year 24x7 support                           HA110A3

Oracle    Oracle Database 10g Release 2 Enterprise Edition,                  4*
          Partitioning, Per Processor              
Oracle    Oracle Database Server Support Package for 3 years                 1
          Incident Server Support -(24x7x4)                        

Oracle    Oracle Application Server 10g - Java Edition, Per Processor        4*
          (v10.1)
Oracle    Oracle Application Server Support Package for 3 years              1
          - Incident Server Support -Package - (24x7x4)
   

(* 4 = 0.50 * 8.  Explanation:  For the purposes of counting the number of
processors which require licensing, an Intel multicore chip with "n" cores
shall be determined by multiplying "n" cores by a factor of 0.50).

  The xerces implementation version 2.6.2 from xml.apache.org was used as it
  comes packaged with Oracle Application Server 10g.
  
  Requests were sent to different Appservers using DNS round
  robin, with the database node acting as the DNS server.
  
  The driver used the following jndi.properties:
  java.naming.factory.initial=com.evermind.server.rmi.RMIInitialContextFactory
  java.naming.provider.url=ormi://main.specj.com:23791/SPECjAppServer
  java.naming.security.principal=oc4jadmin
  java.naming.security.credentials=welcome
  oracle.j2ee.rmi.loadBalance=lookup

The only errors in the driver error logs are the application errors
generated by this benchmark.

On the database, incremental checkpointing ensured that no buffer will
remain dirty (in the cache) for more than the allowed time.

Oracle Application Server 10g Release 10.1.3.2 implements Recoverable Last
Resource Commit optimization where one and only one SinglePhaseResource
participates in a global transaction with one or more XAResources. This is
accomplished by emulating an XAResource to represent the 1PC resource. The
transaction manager takes the following steps when it receives a request to commit
a global transaction that includes one emulated XAResource:
1. It invokes a prepare call on each of the 2PC resources.
2. If all of the 2PC resources are successfully prepared, then commit (one-phase)
is called and the commit record is written to the commit-log, on the emulated XA
resource (database) as part of the local transaction.
3. If the 1PC on the emulated XA resource completes successfully, then the
transaction manager calls commit on each of the 2PC resources.
or:
If the 1PC on the emulated XA resource fails, then the transaction manager queries
the commit-log for transaction outcome. If the record exists, 2PC resources are
committed, otherwise 2PC Resources are aborted.
Transaction outcome after a single point of failure is guaranteed by the presence of
the commit-record in the commit log, indicating a successful transaction completion;
while a missing commit-record signifies abort. The commit record is deleted lazily
after the global transaction has committed.