System Configuration Diagram

  Modifications to the schema where made to enable range partitioning.
  The DDL for creating the range partion tables that were used is
  included in the FDA under the schema/range_part.ddl

  No modifications to the load programs were made.

  The benchmark meets the isolation level requirements by version column
    checking on entities against the database.
  The ItemEnt bean was cached for 20 minute intervals using
    the WebSphere JPA Object Cache mechanism.

  To ensure database durability, RAID10 was used for the
    database logs maintained on the external storage.
    A total of 48 x 36.4GB disks were used for the log.

  Over the course of a 85 minute run at an injection rate of
    4500, the database storage for database system increased by
    10.12 GB.  Given a linear scale, a 24 hour run at the same
    injection rate would increase storage for database system
    by 171.43 GB.  One RAID10 disk arrays were used to create the
    filesystem space for the database.  The array contained
    48 x 36.4GB disks providing 1747 Gb GB of storage.

Supplier  Description                                   Product #          Qty
--------  -------------------------------------------   ----------------   ---
IBM       IBM Power 750 Express                       	8233-E8B             1
-            4x8-core POWER7 Processor
-            (3.55GHz 4MB L3 Cache per core)
-            128GB RAM (32x4GB DIMMS), 1x146.8GB 10k SAS HDD
-            3 Year Onsite Repair 24x7 4 Hour Response
IBM       AIX Standard Edition V6.1 per processor       5765-G62-0103       32
IBM       AIX 3 year support per processor              5773-SM3-1253       32
...
IBM       IBM BladeCenter H Chassis                     7989-BCH             1
IBM       IBM BladeCenter PS702 Express                 8406-71Y + FC 8358   1
-             2x8-core POWER7 Processor
-             (3.0GHz 4MB L3 Cache per core)
-             128GB RAM (32x4GB DIMMS), 1x300GB 10k SAS HDD
-              3 Year Onsite Repair 24x7 4 Hour Response
IBM       DS4800 Disk System Model 82 (4 GB Cache)      1815-82A             1
IBM       DS4000 EXP810 Expansion Unit Model 81         1812-81A             6
IBM       4 Gbps FC, 36.4 GB/15K E-DDM                  5433                96
IBM       AIX Standard Edition V6.1 per processor       5765-G62-0103       16
IBM       AIX 3 year support per processor              5773-SM3-1253       16
...
CDW       3Com Switch 4200G 48-Port                     1020899              2
CDW       10GBASE-CX4 XENPAK Infiniband 4x Module       3CXENPAK95           2
CDW       3Com Switch 4200G 48-Port Sup 1 yr, 24x7x4    819638               4
...
IBM       DB2 Enterprise Server Edition 9.7             D597RLL            16*
-           Lic+SW Maint 12 Months
IBM       DB2 Enterprise Server Edition 9.7             E00BILL             32
-           SW Maint Rewl 1 Anniv
IBM       WebSphere Application Server V7               D55W8LL           32**
-           Lic+SW Maint 12 Months
IBM       WebSphere Application Server V7               E1CBBLL             64
-           SW Maint Rewl 1 Anniv
...
* Note:  Pricing for DB2 is based on Processor Value Units (PVU).  Each POWER7 core is 100 PVU.
** Note:  Pricing for WebSphere is based on Processor Value Units (PVU).  Each POWER7 core is 100 PVU.
** Note:  Apart from WebSphere 7.0.0.9 Application Server, Feature Pack for OSGi Applications and Java Persistence API 2.0 needs to be installed on top

  DNS round robin load balancing was used and hosted on the database machine
  All network connections were 1Gbps

All exceptions in driver logs are from Optimistic Concurrency Checking.
These exceptions are expected in the benchmark.

The IBM DB2 Universal JDBC Drivers is a type 4 driver.

DB2 uses "Soft" checkpoint to ensure that no updates remain unflushed for
longer than the allowed time.

When DB2 UDB Server changes a database table with an update, insert, or delete
operation, the change is initially made in memory, not on disk. When there is
not enough space in the memory buffer to read in or write additional data pages,
DB2 UDB Server will make space by flushing some modified pages to disk.
Modified pages are also written to disk as part of the "Soft" checkpoint to
ensure that no updates remain unflushed for longer than the allowed time.
Before a change is made to the database, it is first recorded in the
transaction log. This ensures that the database can be recovered completely
in the event of a failure. Using the transaction log, transactions that
started but did not complete prior to a failure can be undone, and
transactions recorded as complete in the transaction log but not yet written
to disk can be redone.

DB2 UDB uses a write-ahead-logging protocol to guarantee recovery. This
protocol uses "Soft" checkpoint to write least-recently-used database pages
to disk independent of transaction commit. However, enough log information to
redo/undo the change to a database pages is committed to disk before the
database page itself is written. This protocol therefore renders checkpoint
unnecessary for DB2 UDB. For a more detailed description of the general
principles of the write-ahead-logging protocol, see the IBM research paper,
ARIES: A Transaction Recovery Method Supporting Fine Granularity Locking and
Partial Rollbacks Using Write-Ahead Logging," by C. Mohan, Database Technology
Institute, IBM Almaden Research Center. (http:// portal.acm.org/citation.cfm
?id=128770&coll=portal&dl=ACM&CFID=10343790&CFTOKEN=42047146)