home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
Power Tools 1993 October - Disc 2
/
Power Tools (Disc 2)(October 1993)(HP).iso
/
hotlines
/
cpethl
/
tpcinfo.txt
< prev
Wrap
Text File
|
1992-08-06
|
12KB
|
218 lines
Transaction Processing Performance Council
by Peter Friedenbach, DMSD
Management Summary
The Transaction Processing Performance Council (TPC) is one of several
consortiums working on industry wide performance benchmark standards.
It was founded in 1988 through the initiative of Omri Serlin, an
industry analyst in transaction processing systems. This paper
provides an introduction to the TPC, giving some history behind the
council and explaining the purpose and direction of the future
activities of the council.
Introduction
Originally created in 1988 to address the benchmark abuses centered
around the Debit-Credit and TP1 benchmarks, the TPC has embarked on a
mission to provide a rich set of benchmark suites that will enable a
better environment for competitive performance comparisons across
diverse hardware and software platforms. To this end the council is
nearing the completion of its second formal benchmark specification and
has already laid the foundation for more specifications to follow.
As an organization, the TPC is truly international, made up of 35
member companies representing a wide spectrum of the computer industry.
Everything from big American systems vendors (HP, IBM, Digital, etc.),
major foreign computer companies (NEC, Fijitsu, Hitachi, Bull, etc.),
as well as major database software vendors (Computer Associates,
Oracle, etc.) are members of the council. The council membership
includes the world's ten largest information technology suppliers, and
21 of the top 50, as identified in the Datamation 100 list (Datamation
June 15, 1990).
While the council is currently made up almost entirely of information
technology suppliers, it is not the intention to limit membership to
just vendors. Instead, the council is an open forum and encourages
membership from major customers, consulting firms, and industry
analysts involved in the transaction processing marketplace.
As a product, the council publishes benchmark specifications that
regulate the running and reporting of transaction processing
performance data. It is the goal of each specification to provide a
"level playing field" by which customers are able to make objective
comparisons among performance data published by competing vendors on
different system platforms. Formal implementation of each benchmark is
left to the test sponsor, given the regulations specified within the
standard and the availability of a required full disclosure report.
While it is not a formal requirement, vendors reporting TPC numbers are
strongly urged to have an outside auditor certify the validity of the
performance claims.
TPC Benchmark A
When the council first started, the number one priority was to reach a
standard definition of the perceived "de facto" standard Debit-Credit
benchmark. Up to that point Debit-Credit , and it's derivative TP1,
were the only perceived standards in the transaction processing market.
Unfortunately Debit-Credit was not a real industry standard. The only
public definition of the benchmark appeared in a Datamation article in
1985. In many cases, the benchmark in the article was very loosely
defined. In other circumstances, the definition contained elements
that were overly specific and not readily available across multiple
systems. In general, vendors who published Debit-Credit numbers took
liberties to redefine the benchmark to match the strengths of their
systems with little recourse from the industry and the marketplace.
In November of 1989, the TPC formally published its first benchmark
specification, TPC Benchmark A (TPC-A). TPC-A represents a milestone
in that it is the first time over 30 computer vendors reached agreement
on the basic framework by which OLTP performance measurements should be
conducted.
TPC-A is a system level benchmark that tests the fundamental components
of an OLTP system. While TPC-A is in part derived from the original
Debit-Credit definition, the two benchmarks should not be compared.
TPC-A is a formal benchmark specification, which leaves little room for
vendors to re-interpret the benchmark. While the workload of the
benchmark bears some resemblance to Debit-Credit , much of the work
that went into TPC-A deals with more global issues of what constitutes
legitimate benchmarking practices that will also be applicable to other
specifications that the council may develop in the future.
An often heard complaint about TPC-A is that it is not a representative
benchmark, as anyone who has been associated with its development will
quickly agree. The benchmark should be viewed as a lightweight
reproducible transaction stressing simple elements that are fundamental
components in most transaction processing systems. While the
description of the benchmark has strong overtones of a banking
application, it is only an aid in describing the workload. a real user
application is a much more complex workload made up, in part, of a
series of simple actions tested in TPC-A.
Because TPC-A is not a representative benchmark, it is a mistake to
consider using it for capacity planning. When used properly, TPC-a is
a means of doing a first cut qualification for users looking at
purchasing a new system. To a customer looking for a specific level of
performance, TPC-A will help delineate the wide range of potential
system choices and provide a focused subset that the user can further
refine based upon other selection criteria. TPC-A should never be used
as a substitute for a specific customer application benchmark when
critical capacity planning and/or product evaluation decisions are
involved.
TPC Benchmark B
Having successfully completed work on TPC-A, the council then set out
to address the benchmarking abuses associated with the TP1 benchmark.
The second official specification of the council, TPC Benchmark B (TPC-
B), is the result of this effort and is currently going through formal
approval by the council. If, as expected, two-thirds of the council
members approve the specification, it will become a formal standard.
Completion of the approval process is expected in September 1990.
For the council, TPC-B is an exercise in how to convert a system level
benchmark into a more specific test of the abilities of the database
subsystem. In past TP1 measurements, there was little consistency in
the tricks vendors used to adapt the original Debit-Credit system
benchmark to a more simplistic database test. The result was a set of
performance claims that could not be compared, thrust upon the user
community under the notion of standard TP1 performance measurements.
In deriving TPC-B from TPC-A, the council deliberately spelled out the
exact rules by which vendors can execute and report test results and
therefore, avoid past comparison problems with TP1 measurements.
TPC-B is a database level benchmark that tests the database system
fundamentals common in transaction processing. In essence, TPC-B is a
stress test of the database system's ability to handle transaction
processing. TPC-B performance numbers should never be compared with
the more general system performance of TPC-A. In practice, TPC-B
numbers should be much higher than those of TPC-A and should show
significantly lower price-performance.
As with TPC-A, TPC-B is not a representative benchmark and, therefore,
is not meant for capacity planning. TPC-B should be used instead as a
first cut qualification for users looking to purchase a new database
management system. TPC-B should never be used as a substitute for a
specific customer application benchmark when critical capacity planning
and/or product evaluation decisions are involved.
TPC Workload Spectrum
While TPC-A and TPC-B represent major steps toward true industry
standard benchmarks, the council still recognizes that these benchmarks
represent a limited portion of the entire transaction processing
market. To put direction into its' future standards, the council took
some time to reflect on the requirements of the transaction processing
market and defined a vision for future benchmark development. Out of
this effort has come the definition of the TPC Workload Spectrum.
The TPC Workload Spectrum views the transaction processing benchmark
arena as made up of several different sectors, each representing one or
more benchmarks that address a certain aspect of transaction
processing. The sectors are not meant to represent distinct partitions
of the transaction processing market and in many cases there is
overlap. Rather, the sectors are meant to provide definition and focus
of the activities of the council. Currently the five sectors within
the spectrum are:
Basic System OLTP Services
Business Application Services
Database Stress Tests
Real Time Services
Complex Data Services
The Basic System OLTP Services sector currently encompasses only one
benchmark, TPC-A. As a separate category, TPC-A represents a test of
fundamental elements present in all OLTP applications.
The Business Application Services sector actually represents six
distinct benchmark definitions that, in total, define a set of
application services common to most businesses. Five of the six
benchmarks within the sector are individual workloads focused at
different functional areas of a business (Order Processing, Inventory
Control, Customer Support, Accounting, Decision Support). The sixth
workload will be a composite mix of the previous five. The goal of the
Business Application Services sector is to develop a set of benchmarks,
all based upon a common database, representative of real applications
found in todays corporations.
The Database Stress Tests sector is made up of a set of diverse tests
that exercise different aspects of database management systems.
Currently the council knows of three benchmarks that could be defined
for this sector. TPC-B, which is close to being a formal standard, is
a good test of the transaction processing capabilities of a database
system. The council also sees opportunities for a decision support
benchmark focused on the complex queries found in Executive Information
Systems. A further test exploring the performance of database
utilities and operations is another potential event.
The Real Time Services sector is made up a set of benchmarks, as yet
unspecified, that address the real time requirements of data
acquisition, process control, and message processing. In similar
fashion, Complex Data Services sector also represents a set of
benchmarks that will address the requirements associated with complex
data formats found in records management, document management, and
objects.
With the adoption of the TPC Workload Spectrum, the council has a
vision to guide all future benchmarks. The council is currently
pursuing two specifications within subcommittees. One addresses the
Order Processing portion of the Business Application Services and the
other addresses the Executive Information System portion of the
database stress tests. As time and resources become available in the
future, the council will address the other portions of the spectrum.
Conclusion
It has been almost two years since the TPC was created and with the
pending approval of TPC-B, the council will have finally completed its
original objectives. With the adoption of the TPC Workload spectrum as
a guide, the council has a clear vision of where it would like to take
industry standard transaction processing benchmarks in the future. A
real momentum has developed within the council and the TPC should be
able to deliver on its' promise of a rich set of standard benchmarks
that provide users with a better environment in which to do competitive
performance comparisons of transaction processing systems.