TPC BENCHMARK (TM) H (779138), страница 20
Текст из файла (страница 20)
The sameset of seed values may be used in the consecutive runs.The TPC-H metrics reported for a given system must represent a conservative evaluation of the system’s level ofperformance. Therefore, the reported performance metrics must be for the run with the lower Composite Query-perHour metric5.3.3Power Test5.3.3.1 The power test must be driven by queries submitted by the driver through a single session on the SUT. The sessionexecutes queries one after another. This test is used to measure the raw query execution power of the SUT with asingle query stream. The power test must be executed in parallel with a single refresh stream (see Clause 5.1.2.4).5.3.3.2 The power test must follow these steps in order:1.The refresh function RF1 is executed by the refresh stream.2.The full query set is executed once by the query stream.3.The refresh function RF2 is executed by the refresh stream.5.3.3.3 The timing intervals (see Clause 5.3.7) for each query and for both refresh functions are collected and reported.5.3.4Throughput TestTable 11: Minimum Required Stream CountTPC BenchmarkTM H Standard Specification Revision 2.17.1Page 95SFS(Streams)121033041005300610007300081000093000010100000115.3.4.1 The throughput test must be driven by queries submitted by the driver through two or more sessions on the SUT.There must be one session per query stream on the SUT and each stream must execute queries serially (i.e., one afteranother).
The value of S, the minimum number of query streams, is given in Table 11. The throughput test must beexecuted in parallel with a single refresh stream (see Clause 5.1.2.4).The throughput test must immediately follow one, and only one, power test. No changes to the configuration of theSUT can be made between the power test and the throughput test (see 5.2.7). Any operations performed on the SUTbetween the power and throughput tests must have the following characteristics:They are related to data collection required for the benchmark or requested by the auditorThey are not likely to improve the performance of the throughput test5.3.4.2 When measuring and reporting a throughput test, the number, S, of query streams must remain constant during thewhole measurement interval.
When results are reported with S query streams, these S streams must be the only onesexecuting during the measurement interval (i.e., it is not allowed to execute more than S query streams and reportonly the S best ones).5.3.4.3 For query sequencing purposes (see Clause 5.3.5), each query stream within the throughput test must be assigned aunique stream identification number ranging from 1 to S, the number of query streams in the test.5.3.4.4 When measuring and reporting a throughput test, a single refresh stream (see Clause 5.1.2.4) must be executed inparallel with the S query streams.5.3.5Query Sequencing Rules5.3.5.1 The query sequencing rules apply to each and every query stream, whether part of the power test or part of thethroughput test.5.3.5.2 Each query set has an ordering number, O(s), based on the identification number, s, of the query stream executingthe set.
For example:The query set within the unique query stream of the power test has the ordering number O(00);The query set within the first query stream of the throughput test has the ordering number O(01);The query set within the last of s query streams of the throughput test has the ordering number O(s).TPC BenchmarkTM H Standard Specification Revision 2.17.1Page 965.3.5.3 The sequencing of query executions is done within a query set. The ordering number, O(s), of a query set determinesthe order in which queries must be submitted (i.e., sequenced for execution) within that set and is independent ofany other query set.5.3.5.4 The query submission order of an ordering number, O(s), is given in Appendix A by the ordered set with reference s.Comment: For tests where the list of ordered sets in Appendix A is exhausted, the last reference in the list must befollowed by the first reference in the list (i.e., wrapping around to s = 00).5.3.6Measurement IntervalThe measurement interval, Ts, for the throughput test is measured in seconds as follows:It starts either when the first character of the executable query text of the first query of the first querystream is submitted to the SUT by the driver, or when the first character requesting the execution of thefirst refresh function is submitted to the SUT by the driver, whichever happens first;Comment: In this clause a query stream is said to be first if it starts submitting queries before any other querystreams.It ends either when the last character of output data from the last query of the last query stream is receivedby the driver from the SUT, or when the last transaction of the last refresh function has been completelyand successfully committed at the SUT and a success message has been received by the driver from theSUT, whichever happens last.Comment: In this clause the last query stream is defined to be that query stream whose output data are received lastby the driver.5.3.7Timing Intervals5.3.7.1 Each of the TPC-H queries and refresh functions must be executed in an atomic fashion and timed in seconds.5.3.7.2 The timing interval, QI(i,s), for the execution of the query, Qi, within the query stream, s, must be measuredbetween:The time when the first character of the executable query text is submitted to the SUT by the driver;The time when the first character of the next executable query text is submitted to the SUT by the driver,except for the last query of the set for which it is the time when the last character of the query's output datais received by the driver from the SUT.Comment: All the operations that are part of the execution of a query (e.g., creation and deletion of a temporarytable or a view) must be included in the timing interval of that query.5.3.7.3 The timing interval, RI(j,s), for the execution of the refresh function, RFj, within the refresh stream for the powertest and the throughput test where s is 0 for the power test and s is the position of the pair of refresh functions forthe throughput test, must be measured between:The time when the first character requesting the execution of the refresh function is submitted to the SUTby the driver;The last transaction of the refresh function has been completely and successfully committed at the SUT anda success message has been received by the driver from the SUT.5.3.7.4 The real-time clock used by the driver to compute the timing intervals must be capable of a resolution of at least0.01 second.5.3.7.5 The timing interval of each query and refresh function executed during both tests (i.e., during the power test and thethroughput test) must be individually reported, rounded to the nearest 0.1 second.
For example, 23.74 is rounded to23.7, and 23.75 is rounded to 23.8. Values of less than 0.05 second must be rounded up to 0.1 second to avoid zerovalues.5.3.7.6 The throughput test must include the execution of a single refresh stream. This refresh stream must be used exclusively for the execution of the New Sales refresh function (RF1) and the Old Sales refresh function (RF2).TPC BenchmarkTM H Standard Specification Revision 2.17.1Page 97Comment: The purpose of the refresh stream is to simulate a sequence of batched data modifications executingagainst the database to bring it up to date with its operational data source.5.3.7.7 The refresh stream must execute a number of pairs of refresh functions serially (i.e., one RF1 followed by one RF2)equal to the number of query streams used for the throughput test.Comment: The purpose of this requirement is to maintain a consistent read/write ratio across a wide range of number of query streams.5.3.7.8 The scheduling of each refresh function within the refresh stream is left to the test sponsor with the only requirementthat a given pair must complete before the next pair can be initiated and that within a pair RF1 must complete beforeRF2 can be initiated.Comment: The intent of this Clause is to allow implementations that execute the refresh functions in parallel withthe ad-hoc queries as well as systems that segregate query executions from database refreshes.5.3.7.9 The scheduling of individual refresh functions within an instance of RF1 or RF2 is left to the test sponsor as long asthey meet the requirements of Clause 2.5.2 and Clause 2.5.3.Comment: The intent of this Clause is to allow test sponsors to “trickle” the scheduling of refresh functions tomaintain a more even refresh load throughout the throughput test.5.3.7.10 Prior to the execution of the refresh stream the DBGEN data used for RF1 and RF2 may only be generated, permuted and relocated to the SUT.
![](https://s.studizba.com/z.php?f=/templates/si/i/noavatar.png&w=100&h=100&t=1"){kind=avatar}
