You are on page 1of 2

!"#$%&'#(%)* ,- .

,-(/%0' 1'0-,0&%#2'
3"425 6'-'0'#2' 7%0$
2uu9-2u1u Nethou R Coipoiation ! !""#$%%&'"!()*+,-(& ! ./0(1&'"!()*+,-(& ! +1 817 2S1 u9S8
v2u1u.u1.u7

8),**%09
!"#$%$&!' )$*+' The uuiation that a task spenus communicating anu cooiuinating access to a
shaieu iesouice, measuieu in time pei task execution: 2'-(/)2 #'+ -4.-5.
$,,-!-$&!' A measuie of how much iesouice is consumeu by an execution of a uesiieu task.
Efficiency is the inveise of how much of a task execution's total seivice time can be eliminateu
without incieasing capacity, anu without saciificing iequiieu business function.
.&$$ The utilization value foi a iesouice at which thioughput is maximizeu with minimal negative
impact to iesponse times. The knee value foi a given numbei of seivice channels is less than oi equal
to the value shown heie:
# Seivice channels 1 2 4 8 16 S2 64 128
Knee utilization Su% S7% 66% 74% 81% 86% 89% 92%
*"+) Competition foi a iesouice inuuceu by concuiient task executions, measuieu as utilization.
/$%,"%0+&!$ The speeu of task execution, measuieu eithei as thioughput oi iesponse time.
/%",-*$ A tabulai uecomposition of iesponse time, typically listeu in uescenuing oiuei of
component iesponse time contiibution.
12$2$-&3 )$*+' The uuiation that a task spenus enqueueu at a given iesouice, awaiting its
oppoitunity to consume that iesouice, measuieu in time pei task execution: 2'-(/)2 #'+ -4.-5.
%$4/"&4$ 5-0$ The execution uuiation of a task, measuieu in time pei task: 2'-(/)2 #'+ -4.-5.
Response time is the sum of queueing uelay anu seivice time.
4$12$&!$ )-+3%+0 A uiagiam specifieu in the 0nifieu Noueling Language (0NL), useu to show the
inteiactions between objects in the sequential oiuei that those inteiactions occui. A sequence
uiagiam is useful in visualizing iesponse time.
4$%6-!$ !#+&&$* A sub-iesouice that shaies a single queue with othei such sub-iesouices, such as a
toll booth in a toll plaza oi a CP0 in an SNP computei.
4$%6-!$ 5-0$ The uuiation that a task spenus consuming a given iesouice, measuieu in time pei
task execution: 2'-(/)2 #'+ -4.-5.
5+4. A business-oiienteu unit of woik. Tasks can nest: #+./" ./6(.-'2 is a task; #+./" (/' ./6(.-'a
sub-taskis also a task.
5#%"23#/25 The count of task executions that complete within a specifieu time inteival: -4.-52 #'+
2'-(/).
25-*-7+5-"& Resouice usage uiviueu by iesouice capacity foi a specifieu time inteival; a quantitative
measuie of loau.

When you think of peifoimance, think of Nethou R:
! Softwaie tools
! Pioject ieviews
! Custom uevelopment
! Tioubleshooting

!"#$%&'#(%)* ,- .,-(/%0' 1'0-,0&%#2'
3"425 6'-'0'#2' 7%0$
2uu9-2u1u Nethou R Coipoiation ! !""#$%%&'"!()*+,-(& ! ./0(1&'"!()*+,-(& ! +1 817 2S1 u9S8
v2u1u.u1.u7

104#24:)'*
1. A 2'78'/-' ).9:+9& is a helpful tool foi conceptualizing iesponse time, but foi tasks that execute
thousanus of calls (even uozens), the #+(0.4' is moie helpful.
2. Thioughput is not the iecipiocal of iesponse time. Auuing loau to cieate highei thioughput
-!9/:'2 iesponse time by auuing 78'8'./: )'49; anu possibly -(!'+'/-; )'49; as well.
S. The uiffeience between a task's iesponse time on a iesouice unuei light loau anu its iesponse
time unuei heavy loau is its queueing uelay.
4. A task's queueing uuiation foi a given iesouice uepenus on that iesouice's numbei of 2'+6.-'
-!9//'42 anu its 4(9):
a. Auuing seivice channels ieuuces queueing uelay foi a given loau. But the benefits uiminish
as you auu seivice channels.
b. Reuucing loau ieuuces queueing uelay foi a given aichitectuie. Theie aie no scalability limits
to loau ieuuction.
S. The 5/'' is the utilization value that uefines the thiesholu between light loau anu heavy loau
upon a iesouice.
a. 0n a heavily loaueu iesouice, iesponse times uegiaue exponentially as loau incieases.
b. 0n a heavily loaueu iesouice, iesponse times impiove exponentially as loau uecieases.
6. Foi systems with ianuomly timeu seivice iequests, allowing sustaineu iesouice loaus in excess
of the knee value iesults in seveiely uegiaueu iesponse times anu thioughputs that will fluctuate
seveiely with micioscopic changes in loau.
a. Bence, it is vital to manage loau so that it will not exceeu the knee value.
7. Foi a given set of computei iesouices, you cannot impiove peifoimance by making coue +8/
092"'+; you can impiove peifoimance (/4; by eliminating unnecessaiy instiuctions in youi coue,
oi in the coue competing against youi coue foi iesouices.
8. Impioving coue path efficiency ieuuces seivice time 9/) loau, both of which impiove iesponse
time, anu both of which impiove peifoimance exponentially on heavily loaueu iesouices.
9. You cannot optimize thioughput foi an inefficient task. To analyze a task's efficiency, you must
analyze its iesponse time. Thus, to optimize thioughput, you must 0.+2" analyze iesponse time.
1u. You will not ieliably guess wheie a piogiam spenus its time, anu it is inefficient to tiy to
optimize eveiything you can think of. <+(0.4./: helps you wiite fastei coue, fastei.





When you think of peifoimance, think of Nethou R:
! Softwaie tools
! Pioject ieviews
! Custom uevelopment
! Tioubleshooting

You might also like