tpcc_current (1034761), страница 17
Текст из файла (страница 17)
999] for the first1,000 cu stom ers, and generating a non -u niform rand om nu m ber u sing the fu nctionN URand (255,0,999) for each of the rem aining 2,000 custom ers. The ru n -tim e constant C (see Clau se2.1.6) u sed for the d atabase p op u lation m u st be rand om ly chosen ind ep end ently from the testru n(s).C_MIDDLE = "OE"C_FIRST rand om a-string [8 .. 16]C_STREET_1 rand om a-string [10 ..
20]C_STREET_2 rand om a-string [10 .. 20]C_CITY rand om a-string [10 .. 20]C_STATE rand om a-string of 2 lettersC_ZIP generated accord ing to Clau se 4.3.2.7C_PH ON E rand om n-string of 16 nu m bersC_SIN CE d ate/ tim e given by the op erating system w hen the CUSTOMER table w as p op u lated .C_CREDIT = "GC". For 10% of the row s, selected at rand om , C_CREDIT = "BC"C_CREDIT_LIM = 50,000.00C_DISCOUN T rand om w ithin [0.0000 .. 0.5000]C_BALAN CE = -10.00C_YTD_PAYMEN T = 10.00C_PAYMEN T_CN T = 1C_DELIVERY_CN T = 0C_DATA rand om a-string [300 .. 500]For each row in the CUSTOMER table:- 1 row in the H ISTORY table w ith:H _C_ID = C_IDH _C_D_ID = H _D_ID = D_IDH _C_W_ID = H _W_ID = W_IDH _DATE cu rrent d ate and tim eH _AMOUN T = 10.00H _DATA rand om a-string [12 .. 24]TPC Benchm ark™C - Stand ard Sp ecification, Revision 5.11 - Page 67 of 130* 3,000 row s in the ORDER table w ith:O_ID u niqu e w ithin [3,000]O_C_ID selected sequ entially from a rand om p erm u tation of [1 ..
3,000]O_D_ID = D_IDO_W_ID = W_IDO_EN TRY_D cu rrent d ate/ tim e given by the op erating systemO_CARRIER_ID rand om w ithin [1 .. 10] if O_ID < 2,101, nu ll otherw iseO_OL_CN T rand om w ithin [5 .. 15]O_ALL_LOCAL = 1For each row in the ORDER table:- A nu m ber of row s in the ORDER-LIN E table equ al to O_OL_CN T, generated accord ing to the ru lesfor inp u t d ata generation of the N ew -Ord er transaction (see Clau se 2.4.1) w ith:OL_O_ID = O_IDOL_D_ID = D_IDOL_W_ID = W_IDOL_N UMBER u niqu e w ithin [O_OL_CN T]OL_I_ID rand om w ithin [1 .. 100,000]OL_SUPPLY_W_ID = W_IDOL_DELIVERY_D = O_ENTRY_D if OL_O_ID < 2,101, nu ll otherw iseOL_QUAN TITY = 5OL_AMOUN T = 0.00 if OL_O_ID < 2,101, rand om w ithin [0.01 ..
9,999.99] otherwiseOL_DIST_IN FO rand om a-string of 24 letters* 900 row s in the N EW-ORDER table corresp ond ing to the last 900 row s in the ORDER table for thatd istrict (i.e., w ith N O_O_ID betw een 2,101 and 3,000), w ith:N O_O_ID = O_IDN O_D_ID = D_IDN O_W_ID = W_IDComment: Five p ercent (5%) variation from the target card inality of S_DATA w ith ” ORGIN AL” , I_DATA w ith“ ORIGIN AL” , and C_CREDIT w ith “ BC” is allow ed to accou nt for the rand om variation encou ntered d u ring theinitial d ata load ing of the d atabase.4.3.3.2The im p lem entation m ay not take ad vantage of the fact that som e field s are initially p op u lated w ith afixed valu e.
For exam p le, storage sp ace cannot be saved by d efining a d efau lt valu e for the field C_CREDIT_LIMand storing this valu e only once in the d atabase.TPC Benchm ark™C - Stand ard Sp ecification, Revision 5.11 - Page 68 of 130Clause 5: PERFORMAN CE METRICS and RESPON SE TIME5.1D efinition of Terms5.1.1The term measurement interval refers to a stead y state p eriod d u ring the execu tion of the benchm arkfor w hich the test sp onsor is rep orting a throu ghp u t rating (see Clau se 5.5 for d etailed requ irem ents).5.1.2The term completed transactions refers to any bu siness transaction (see Clau se 2.1.3) that has beensu ccessfu lly com m itted at the SUT and w hose ou tp u t d ata has been d isp layed by the Rem ote Term inal Em u lator (incase of a N ew -Ord er, Paym ent, Ord er-Statu s, or Stock-Level transaction ) or for w hich a com p lete entry has beenw ritten into a resu lt file (in case of a Delivery transaction ).
N ew -Ord er transaction s that are rolled back, as requ iredby Clau se 2.4.1.4, are consid ered as com p leted transactions.5.2Pacing of Transactions by Emulated Users5.2.1The figu re below illu strates the cycle execu ted by each em u lated u ser (see Clau se 5.2.2). The activep ortion of the screen is rep resented w ith bold face text:Prev iousScreenmenu1 - Select transaction ty pe3 - Measure Menu RT2 - Display ScreenInputScreenmenu4 - Wait (Key ing Time)6 - Measure Txn. RTOutputScreenmenu5 - Display Data7 - Wait (Think Time)5.2.2follow s:1.Each em u lated u ser execu tes a cycle com p rised of screens, w ait tim es, and resp onse tim es (RTs) asSelects a transaction typ e from the m enu accord ing to a w eighted d istribu tion (see Clau se 5.2.3).TPC Benchm ark™C - Stand ard Sp ecification, Revision 5.11 - Page 69 of 1302.Waits for the Inp u t/ Ou tp u t Screen to be d isp layed .3.Measu res the Menu RT (see Clau se 5.3.3).4.Enters the requ ired num ber of inp u t field s (see Clau se 2) over the d efined m inim u m Keying Tim e (seeClau se 5.2.5.2).5.Waits for the requ ired nu m ber of ou tp u t field s (see Clau se 2) to be d isp layed on the Inp u t/ Ou tp u t Screen.6.Measu res the Transaction RT (see Clau se 5.3.4).7.Waits for the d efined m inim u m Think Tim e (see Clau se 5.2.5.4) w hile the inp ut/ ou tp u t screen rem ainsd isp layed .At the end of the Think Tim e (Step 7) the em u lated u ser loop s back to select a transaction typ e from the m enu (Step1).Comment: N o action is requ ired on the p art of the SUT to cycle from Step 7 back to Step 1.5.2.3Each transaction typ e (i.e., bu siness transaction ) is available to each term inal throu gh the Menu .
Overthe m easu rem ent interval, the term inal p op u lation m u st m aintain a m inim u m p ercentage of m ix for each transactiontyp e as follow s:Transaction Typ eN ew -Ord er 1Paym entOrd er-Statu sDeliveryStock-LevelMinim u m % of m ixn/ a43.04.04.04.01 There is no m inim u m for the N ew -Ord er transaction as its m easu red rate is the rep orted throu ghp u t .Comment 1: The intent of the m inim um p ercentage of m ix for each transaction typ e is to execu te ap p roxim ately onePaym ent transaction for each N ew -Ord er transaction and ap p roxim ately one Ord er-Statu s transaction , one Deliverytransaction , and one Stock-Level transaction for every 10 N ew -Ord er transactions.
This m ix resu lts in the com p letebu siness p rocessing of each ord er.Comment 2: The total nu m ber of transactions, from w hich the m inim u m p ercentages of m ix are d erived , m ay becalcu lated in either of tw o w ays:Based on all transactions that w ere selected from the Menu and com p leted (see Clau se 5.1.2) w ithin them easu rem ent interval.Based on all tran sactions w hose Transaction RT (see Clau se 5.3.4) w as com p letely m easu red at the RTEd u ring the m easu rem ent interval.Comment 3: As an ease of benchm arking issu e, the ap p roach in Clau se 5.4.2 m ay be u sed to com p u te the transactionm ix p ercentage and throu gh p u t d ata.TPC Benchm ark™C - Stand ard Sp ecification, Revision 5.11 - Page 70 of 1305.2.4Regulation of Transaction MixTransaction typ es m u st be selected u niform ly at rand om w hile m aintaining the requ ired m inim u m p ercentage ofm ix for each transaction typ e over the m easu rem ent interval.
This m u st be d one u sing one of the techniqu esd escribed in Clau ses 5.2.4.1 and 5.2.4.2.5.2.4.1A w eight is associated to each transaction typ e on the m enu . The requ ired m ix is achieved by selectingeach new transaction u niform ly at rand om from a w eighted d istribu tion. The follow ing requ irem ents m u st besatisfied w hen u sing this techniqu e:1.The actu al w eights are chosen by the test sp onsor and m u st resu lt in m eeting the requ ired m inim u mp ercentages of m ix in Clau se 5.2.3.2.For the p u rp ose of achieving the requ ired transaction m ix, the RTE can d ynam ically ad ju st the w eigh tassociated to each transaction typ e d u ring the m easu rem ent interval. These ad ju stm ents m u st be lim itedso as to keep the w eights w ithin 5% on either sid e of their resp ective initial valu e.5.2.4.2One or m ore card s in a d eck are associated to each transaction typ e on the Menu .
The requ ired m ix isachieved by selecting each new transaction u niform ly at rand om from a d eck w hose content gu arantees the requ iredtransaction m ix. The follow ing requ irem ents m u st be satisfied w hen u sing this techniqu e:1.Any nu m ber of term inals can share the sam e d eck (inclu d ing bu t not lim ited to one d eck p er term inal orone d eck for all term inals).2.A d eck m u st be com p rised of one or m ore sets of 23 card s (i.e., 10 N ew -Ord er card s, 10 Paym ent card s,and one card each for Ord er-Statu s, Delivery, and Stock-level).
The m inim u m size of a d eck is one set p erterm inal sharing this d eck. If m ore than one d eck is u sed , then all d ecks m u st be of equ al sizes.Comment: Generating the m axim u m p ercentage of N ew -Ord er transaction s w hile achieving the requ iredm ix can be d one for exam p le by sharing a d eck of 230 card s betw een 10 term inals.3.Each p ass throu gh a d eck m u st be m ad e in a d ifferent u niform ly rand om ord er.
If a d eck is accessedsequ entially, it m u st be rand om ly shu ffled each tim e it is exhau sted . If a d eck is accessed at rand om , card sthat are selected cannot be p laced back in the d eck u ntil it is exhau sted .Comment: All term inals m u st select transactions u sing the sam e techniqu e. Gaining a p erform ance or ap rice/ p erform ance ad vantage by d riving one or m ore term inals d ifferently than the rest of the term inal p op u lationis not allow ed .5.2.5Wait Times and Response Time Constraints5.2.5.1The Menu step is transaction ind ep end ent. At least 90% of all Menu selections m u st have a Menu RT(see Clau se 5.3.3) of less than 2 second s.5.2.5.2For each transaction typ e, the Keying Tim e is constant and m u st be a m inim u m of 18 second s for N ew Ord er, 3 second s for Paym ent, and 2 second s each for Ord er-Statu s, Delivery, and Stock-Level.5.2.5.3At least 90% of all transactions of each typ e m u st have a Transaction RT (see Clau se 5.3.4) of less than5 second s each for N ew -Ord er, Paym ent, Ord er-Statu s, and Delivery, and 20 second s for Stock-Level.Comment: The total nu m ber of transactions, from w hich the Transaction RT of N ew -Ord er is com p u ted , inclu d esN ew -Ord er transaction s that rollback as requ ired by Clau se 2.4.1.4.TPC Benchm ark™C - Stand ard Sp ecification, Revision 5.11 - Page 71 of 1305.2.5.4For each transaction typ e, think tim e is taken ind ep end ently from a negative exp onential d istribu tion.Think tim e, Tt , is com p u ted from the follow ing equ ation:Tt = -log(r) * w here: log = natu ral log (base e)Tt = think tim er = rand om nu m ber uniform ly d istribu ted betw een 0 and 1 = m ean think tim eEach d istribu tion m ay be tru ncated at 10 tim es its m ean valu e5.2.5.5The beginning of all w ait tim es (Keying Tim es and Think Tim es) are to be taken after the last characterof ou tp u t has been d isp layed (see Clau se 2.2.2) by the em u lated term inal.5.2.5.6The 90th p ercentile resp onse tim e for the N ew -Ord er, Paym ent, Ord er-Statu s, Stock-Level and theinteractive p ortion of the Delivery transactions m u st be greater than or equ al to the average resp onse tim e of thattransaction.