| Re. .2
> - has TPC_C been approved by the council?
Yes, as of early August, TPC-C has been approved by the Council.
The final draft of the specification is ready; I should have a copy
in a few days.
> - what were the AS/400 numbers?
System tpmC $/tpmC
------ ---- ------
AS/400 9404 E10 33.81 2,462
AS/400 9406 E35 54.14 3,483
> I didn't see anything in the August 28th version of all TPC results
> from TPSDOC::sys$public:[tp_performance]tpc_summary.txt
Probably because the IBM results were submitted Sept. 2.
|
| I was asked to post this basic info on TPC-C here. If you are interested in
more details, do copy the overview paper referred to at the end. If you need
even more details, drop me a line.
-amitabh.
------------------------------------------------------------------------------
What is the TPC-C Benchmark?
----------------------------
by
Amitabh Shah and Walt Kohler
TNSG Software Performance Group
September 18, 1992
In 1988, Digital's Database Performance Group conducted a technical survey of
existing database and transaction processing (TP) workloads for the purpose of
standardizing its own internal performance testing. Among the workloads
examined were: Avertz (a Digital internal workload), DebitCredit (the already
emerging industry standard), the Wisconsin benchmark, MCC's Order-Entry
benchmark, RAMP-C, and several other Digital internal workloads. The MCC
Order-Entry workload was selected as the basis for Digital's internal
benchmark because of its technology focus and wide applicability.
The Order-Entry workload was extensively modified by Digital; while the basic
concepts introduced by MCC were kept, the database and the transactions were
remodeled. The resulting workload was written into benchmark proposal and
presented to the Transaction Processing Performance Council (TPC) in
March 1990 for standardization. The TPC accepted the proposal in May 1990
and formed a subcommittee to refine it in July 1990. The subcommittee
consisted of representatives from Amdahl, Bull H.N., Control Data, Data
General, Fujitsu/ICL, Hewlett-Packard, IBM, Informix, Mips Computer,
Oracle, Sequent Computer, Sun, Sybase, Tandem, and Unisys, in addition to
Digital. The TPC-C benchmark was finalized in June 1992, and was approved as a
Standard by the TPC in August 1992.
In many fundamental areas, TPC-C carries over the characteristics of
TPC-A, the TPC's standard version of DebitCredit. As such, TPC-C
- is a multi-user benchmark that requires a Remote Terminal Emulator to
emulate a population of users and their terminals
- uses a database that scales with the throughput of the System Under Test
- requires support of the ACID properties of transactions
- requires a Full Disclosure Report
- reports performance (transaction throughput) and price/performance (system
price/throughput) metrics.
But unlike TPC-A, the new benchmark is designed to cover aspects of
real-life systems that are not found in the simpler DebitCredit model,
such as
- more complex database structure
- multiple transaction types of varying complexity
- online and deferred execution modes
- higher levels of contention on data access and update
- data access by primary as well as non-primary keys
- non-uniform data access patterns to simulate data hot-spots
- realistic requirements for terminal input/output with full screen
formatting
- requirements for full transparency of data partitioning
- transaction rollbacks.
The workload models the activity of a wholesale supplier with a number of
geographically distributed sales districts and associated supply warehouses.
The workload transactions model the activity of customers placing orders,
making payments for their orders, or simply inquiring about their orders. In
addition, there are also transactions that model customers' orders being
delivered, as well as in-house inventory check operations.
The realism and the complexity of TPC-C will make it a better choice for
benchmarking OLTP applications. It is expected that TPC-C will replace TPC-A
as the predominant TP benchmark in the future.
The TNSG Software Performance Group (SPG) represents Digital at the TPC and
has implemented a version of TPC-C for VAX/VMS systems. Walt Kohler
is Digital's primary representative to the TPC; he chaired the TPC
subcommittee that developed the TPC-C benchmark. Amitabh Shah was a member
of the TPC-C development subcommittee, and the project leader responsible
for TPC-C development and testing at Digital during the standardization
process. The benchmark has been implemented using Rdb/VMS V4.2, ACMS V3.3,
and DECforms V1.4. Tuning of the application for performance currently is
under way by Yongmin Chen. It is expected that SPG's implementation of the
TPC-C workload will provide a good system level test for VMS and many
important NAS software products and services.
A more comprehensive description of the TPC-C benchmark is contained in the
report entitle "Overview of TPC Benchmark C" by Walt Kohler and Amitabh
Shah. The postscript file for this report is available by typing:
$ COPY tpsdoc::sys$public:[tp_performance.tpc]tpc_c_overview.ps *.*
For more information on TPC-C and Digital's TPC performance results,
contact Hwan Shen (TPSYS::SHEN, DTN 227-4426).
|