tpcc_current (1034761), страница 18
Текст из файла (страница 18)
If the 90th and the average resp onse tim es are d ifferent by less that 100m s (.1 second s), then they areconsid ered equ al. This requ irem ent is for the term inal resp onse tim es only and d oes not ap p ly to the d eferredp ortion of the Delivery transaction or to the m enu step .5.2.5.7The follow ing table su m m arizes the tran saction m ix, wait tim es, and resp onse tim e constraints:TransactionTyp eMinim u m% of m ixN ew -Ord erPaym entOrd er-Statu sDelivery 1Stock-Leveln/ a43.04.04.04.0Minim u mKeying Tim e18 sec.3 sec.2 sec.2 sec.2 sec.90th PercentileResp onse Tim eConstraintMinim u m Meanof Think Tim eDistribu tion5 sec.5 sec.5 sec.5 sec.20 sec.12 sec.12 sec.10 sec.5 sec.5 sec.1 The resp onse tim e is for the term inal resp onse (acknow led ging that the transaction has been qu eu ed ), notfor the execu tion of the transaction itself.
At least 90% of the transactions m u st com p lete w ithin 80 second sof their being qu eu ed (see Clau se 2.7.2.2).Comment 1: The resp onse tim e constraints are set su ch that the throu ghp u t of the system is exp ected to beconstrained by the resp onse tim e requ irem ent for the N ew -Ord er transaction .
Resp onse tim e constraints for othertransactions are relaxed for that p u rp ose.Comment 2: The keying tim es for the transactions are chosen to be ap p roxim ately p rop ortional to the nu m ber ofcharacters inp u t, and the think tim es are chosen to be ap p roxim ately p rop ortional to the nu m ber of charact ersou tp u t.5.2.5.8For each transaction typ e, all configu red term inals of the tested system s m u st u se the sam e targetKeying Tim e and the sam e target m ean of Think Tim e.
These tim es m u st com p ly w ith the requ irem ents su m m arizedin Clau se 5.2.5.7.TPC Benchm ark™C - Stand ard Sp ecification, Revision 5.11 - Page 72 of 1305.3Response Time D efinition5.3.1Each com p leted transaction su bm itted to the SUT m u st be ind ivid u ally tim ed .5.3.2Resp onse Tim es m u st be m easu red at the RTE. A Response Time (or RT) is d efined by:RT = T2 - T1w here:T1 and T2 are m easu red at the RTE and d efined as:T1 = tim estam p taken before the last character of inp u t d ata is entered by the em u lated u s er.T2 = tim estam p taken after the last character of ou tp u t is received by the em u lated term inal.The resolu tion of the tim estam p s m u st be at least 0.1 second s.Comment: The intent of the benchm ark is to m easu re resp onse tim e as exp erienced by the em u lated u ser.5.3.3The Menu Resp onse Tim e (Menu RT) is the tim e betw een the tim estam p taken before the lastcharacter of the Menu selection has been entered and the tim estam p taken after th e last character of theInp u t/ Ou tp u t Screen has been received (inclu d ing clearing all inp u t and ou tp u t field s and d isp laying fixed field s,see Clau se 2).Comment: System s that d o not requ ire SUT/ RTE interaction for the Menu selection and the screen d isp lay canassu m e a nu ll Menu RT and the com p onents that p rovid e the resp onse for the Menu requ est (e.g.
screen cachingterm inals) m u st be inclu d ed in the SUT and therefore m u st be p riced .5.3.4The Transaction Resp onse Tim e (Transaction RT) is the tim e betw een the tim estam p taken before thelast character of the requ ired inp u t d ata has been sent from the RTE (see Clau se 2) and the tim estam p taken after thelast character of the requ ired ou tp u t d ata has been received by the RTE (see Clau se 2) resu lting from a transactionexecu tion.Comment: If the em u lated term inal m u st p rocess the d ata being entered or d isp layed , the tim e for this p rocessingm u st be d isclosed and taken into accou nt w hen calcu lating the Transaction RT.5.4Computation of Throughput RatingThe TPC-C transaction m ix rep resents a com p lete bu siness cycle. It consists of m u ltip le bu siness transaction s w hichenter new ord ers, qu ery the statu s of existing ord ers, d eliver ou tstand ing ord ers, enter p aym ents from custom ers,and m onitor w arehou se stock levels.5.4.1The m etric u sed to rep ort Maxim u m Qu alified Throu ghp u t (MQTh) is a nu m ber of ord ers p rocessedp er m inu te.
It is a m easu re of "bu siness throu ghp u t " rather than a transaction execu tion rate. It im p licitly takes intoaccou nt all transactions in the m ix as their ind ivid u al throu ghp u t is controlled by the w eighted Menu selection andthe m inim u m p ercentages of m ix d efined in Clau se 5.2.3.TPC Benchm ark™C - Stand ard Sp ecification, Revision 5.11 - Page 73 of 1305.4.2The rep orted MQTh is the total nu m ber of com p leted N ew -Ord er transaction s (see Clau se 5.1.2),w here the Transaction RT (see Clau se 5.3.4) w as com pletely m easu red at the RTE d u ring the m easu rem ent interval,d ivid ed by the elap sed tim e of the interval. N ew -Ord er transactions that rollback, as requ ired by Clau se 2.4.1.4, m u stbe inclu d ed in the rep orted MQTh.5.4.3The nam e of the m etric u sed to rep ort the MQTh of the SUT is tpmC.5.4.4All rep orted MQTh m u st be m easu red , rather than interp ola ted or extrap olated , and truncated toexactly zero d ecim al p laces.
For exam p le, su p p ose 105.548 tp m C is m easu red on a 100 term inal test for w hich 90% ofthe N ew -Ord er transactions com p leted in less than 4.8 second s and 117.572 tp m C is m easu red on a 110 term inal testfor w hich 90% of the transactions com p leted in less than 5.2 second s. Then the reported tp m C is 105.5.4.5To be valid , the m easu rem ent interval m u st contain no m ore than 1% or no m ore than one (1),w hichever is greater, of the Delivery transaction s skip p ed becau se there w ere few er than necessary ord ers p resent inthe N ew -Ord er table.5.5Measurement Interval Requirements5.5.1Steady State5.5.1.1of the SUT.The test m u st be cond u cted in a steady state cond ition that rep resents the tru e su stainable throu ghp u t5.5.1.2Althou gh the m easu rem ent interval m ay be as short as 120 m inu tes, the system u nd er test m u st beconfigu red to ru n the test at the rep orted tp m C for a continu ou s p eriod of at least eight hou rs w ithou t op eratorintervention , m aintaining fu ll ACID p rop erties.
For exam p le, the m ed ia u sed to store at least 8 hou rs of log d atam u st be configu red if requ ired to recover from any single p oint of failu re (see Clau se 3.5.3.1).Comment 1: An exam p le of a configu ration that w ou ld not com p ly is one w here a log file is allocated su ch thatbetter p erform ance is achieved d u ring the m easu red portion of the test than d u ring the rem aining p ortion of an eighthou r test, p erhap s becau se a d ed icated d evice w as u sed initially bu t sp ace on a shared d evice is u sed later in the fu lleight hou r test.Comment 2: Stead y state is easy to d efine (e.g., su stainable throu ghp u t ) bu t d ifficu lt to p rove. The test sp onsor(and / or the au d itor) is requ ired to rep ort the m ethod u sed to verify stead y state su stainable throu ghp u t.
The au d itoris encou raged to u se available m onitoring tools to help d eterm ine the stead y state.Comment 3: Som e asp ects of an im p lem entation can resu lt in system atic bu t sm all va riations in su stainedthrou ghp u t over an 8 hou r p eriod . The cu m u lative effect of su ch variations m ay be u p to 2% of the rep ortedthrou ghp u t. There is no requ irem ent for an 8 hou r ru n.5.5.1.3In the case w here a ram p -u p p eriod is u sed to reach stead y state, the p rop erly scaled initial d atabasep op u lation is requ ired at the beginning of the ram p u p p eriod . The transaction m ix and the requ irem entssu m m arized in Clau se 5.2.5.7 m u st be follow ed d u ring the ram p -u p as w ell as stead y state p eriod .Comment: The intent of this clau se is to p revent significant alteration to the p rop erly scaled initial d atabasep op u lation d u ring the ram p -u p p eriod .5.5.1.4A sep arate m easu rem ent to d em onstrate rep rod u cibility is not requ ired .TPC Benchm ark™C - Stand ard Sp ecification, Revision 5.11 - Page 74 of 1305.5.1.5While variability is allow ed , the RTE cannot be artificially w eighted to generate inp u t d ata d ifferentfrom the requ irem ents d escribed in Clau ses 2.4.1, 2.5.1, 2.6.1, 2.7.1, and 2.8.1.
To be valid , the inp u t d ata generat edd u ring a rep orted m easu rem ent interval m u st not exceed the follow ing variability:1.At least 0.9% and at m ost 1.1% of the N ew -Ord er transaction s m u st roll back as a resu lt of an u nu sed itemnu m ber.2.The average num ber of ord er-lines p er ord er m u st be in the range of 9.5 to 10.5 and the num ber of ord er lines p er ord er m u st be u niform ly d istribu ted from 5 to 15 for the N ew -Ord er transaction s that aresu bm itted to th e SUT d u ring the m easu rem ent interval.3.The nu m ber of rem ote ord er-lines m u st be at least 0.95% and at m ost 1.05% of the nu m ber of ord er -linesthat are filled in by the N ew -Ord er transactions that are su bm itted to the SUT d u ring the m easu rem entinterval.4.The num ber of rem ote Paym ent transaction s m u st be at least 14% and at m ost 16% of the nu m ber ofPaym ent transactions that are su bm itted to the SUT d u ring the m easu rem ent interval.5.The nu m ber of cu stom er selections by cu stom er last nam e in the Paym ent transaction m u st be at least 57%and at m ost 63% of the nu m ber of Paym ent transactions that are su bm itted to the SUT d u ring them easu rem ent interval.6.The nu m ber of cu stom er selections by cu stom er last nam e in the Ord er-Statu s transaction m u st be at least57% and at m ost 63% of the nu m ber of Ord er -Statu s transactions that are su bm itted to the SUT d u ring them easu rem ent interval.5.5.1.6To be valid , the m easu rem ent interval m u st contain no m ore than 1% or no m ore than one (1),w hichever is greater, of the Delivery transaction s skip p ed becau se there w ere few er than necessary ord ers p resent inthe N ew -Ord er table.5.5.25.5.2.1D urationThe m easu rem ent interval m u st:1.Begin after the system reaches stead y state.2.Be long enou gh to generate rep rod u cible th rou ghp u t resu lts w hich are rep resentative of the p erform ancew hich w ou ld be achieved d u ring a su stained eight hou r p eriod .3.Extend u ninterru p ted for a m inim u m of 120 m inu tes.5.5.2.2Som e system s d o not w rite m od ified d atabase record s/ p ages to d u rable m ed ia at the tim e ofm od ification, bu t instead d efer these w rites.