<s>
In	O
multiprocessor	B-Operating_System
computer	O
systems	O
,	O
software	B-Operating_System
lockout	I-Operating_System
is	O
the	O
issue	O
of	O
performance	O
degradation	O
due	O
to	O
the	O
idle	O
wait	O
times	O
spent	O
by	O
the	O
CPUs	B-Device
in	O
kernel-level	O
critical	B-Operating_System
sections	I-Operating_System
.	O
</s>
<s>
Software	B-Operating_System
lockout	I-Operating_System
is	O
the	O
major	O
cause	O
of	O
scalability	B-Architecture
degradation	O
in	O
a	O
multiprocessor	B-Operating_System
system	O
,	O
posing	O
a	O
limit	O
on	O
the	O
maximum	O
useful	O
number	O
of	O
processors	O
.	O
</s>
<s>
To	O
mitigate	O
the	O
phenomenon	O
,	O
the	O
kernel	B-Operating_System
must	O
be	O
designed	O
to	O
have	O
its	O
critical	B-Operating_System
sections	I-Operating_System
as	O
short	O
as	O
possible	O
,	O
therefore	O
decomposing	O
each	O
data	B-General_Concept
structure	I-General_Concept
in	O
smaller	O
substructures	O
.	O
</s>
<s>
In	O
most	O
multiprocessor	B-Operating_System
systems	O
,	O
each	O
processor	B-Device
schedules	O
and	O
controls	O
itself	O
,	O
therefore	O
there	O
's	O
no	O
"	O
supervisor	O
"	O
processor	B-Device
,	O
and	O
kernel	B-Operating_System
data	B-General_Concept
structures	I-General_Concept
are	O
globally	O
shared	O
;	O
sections	O
of	O
code	O
that	O
access	O
those	O
shared	O
data	B-General_Concept
structures	I-General_Concept
are	O
critical	B-Operating_System
sections	I-Operating_System
.	O
</s>
<s>
This	O
design	O
choice	O
is	O
made	O
to	O
improve	O
scaling	B-Architecture
,	O
reliability	O
and	O
modularity	O
.	O
</s>
<s>
Examples	O
of	O
such	O
kernel	B-Operating_System
data	B-General_Concept
structure	I-General_Concept
are	O
ready	O
list	O
and	O
communication	O
channels	O
.	O
</s>
<s>
A	O
"	O
conflict	O
"	O
happens	O
when	O
more	O
than	O
one	O
processor	B-Device
is	O
trying	O
to	O
access	O
the	O
same	O
resource	O
(	O
a	O
memory	O
portion	O
)	O
at	O
the	O
same	O
time	O
.	O
</s>
<s>
To	O
prevent	O
critical	O
races	O
and	O
inconsistency	O
,	O
only	O
one	O
processor	B-Device
(	O
CPU	B-Device
)	O
at	O
a	O
given	O
time	O
is	O
allowed	O
to	O
access	O
a	O
particular	O
data	B-General_Concept
structure	I-General_Concept
(	O
a	O
memory	O
portion	O
)	O
,	O
while	O
other	O
CPUs	B-Device
trying	O
to	O
access	O
at	O
the	O
same	O
time	O
are	O
locked-out	B-Operating_System
,	O
waiting	O
in	O
idle	O
status	O
.	O
</s>
<s>
The	O
idle	O
wait	O
is	O
not	O
necessary	O
but	O
convenient	O
in	O
the	O
case	O
of	O
a	O
critical	B-Operating_System
section	I-Operating_System
for	O
synchronization/IPC	O
operations	O
,	O
which	O
require	O
less	O
time	O
than	O
a	O
context	B-Operating_System
switch	I-Operating_System
(	O
executing	O
another	O
process	B-Operating_System
to	O
avoid	O
idle	O
wait	O
)	O
.	O
</s>
<s>
Idle	O
wait	O
is	O
instead	O
not	O
convenient	O
in	O
case	O
of	O
a	O
kernel	B-Operating_System
critical	B-Operating_System
section	I-Operating_System
for	O
device	O
management	O
,	O
present	O
in	O
monolithic	B-Operating_System
kernels	I-Operating_System
only	O
.	O
</s>
<s>
A	O
microkernel	B-Operating_System
instead	O
falls	O
on	O
just	O
the	O
first	O
two	O
of	O
the	O
above	O
cases	O
.	O
</s>
<s>
In	O
a	O
multiprocessor	B-Operating_System
system	O
,	O
most	O
of	O
the	O
conflicts	O
are	O
kernel-level	O
conflicts	O
,	O
due	O
to	O
the	O
access	O
to	O
the	O
kernel	B-Operating_System
level	O
critical	B-Operating_System
sections	I-Operating_System
,	O
and	O
thus	O
the	O
idle	O
wait	O
periods	O
generated	O
by	O
them	O
have	O
a	O
major	O
impact	O
in	O
performance	O
degradation	O
.	O
</s>
<s>
This	O
idle	O
wait	O
time	O
increases	O
the	O
average	O
number	O
of	O
idle	O
processors	O
and	O
thus	O
decreases	O
scalability	B-Architecture
and	O
relative	O
efficiency	O
.	O
</s>
<s>
Taking	O
as	O
parameters	O
the	O
average	O
time	O
interval	O
spent	O
by	O
a	O
processor	B-Device
in	O
kernel	B-Operating_System
level	O
critical	B-Operating_System
sections	I-Operating_System
(	O
L	O
,	O
for	O
time	O
in	O
locked	O
state	O
)	O
,	O
and	O
the	O
average	O
time	O
interval	O
spent	O
by	O
a	O
processor	B-Device
in	O
tasks	O
outside	O
critical	B-Operating_System
sections	I-Operating_System
(	O
E	O
)	O
,	O
the	O
ratio	O
L/E	O
is	O
crucial	O
in	O
evaluating	O
software	B-Operating_System
lockout	I-Operating_System
.	O
</s>
<s>
In	O
a	O
system	O
with	O
a	O
L/E	O
ratio	O
of	O
0.05	O
,	O
for	O
instance	O
,	O
if	O
there	O
are	O
15	O
CPUs	B-Device
,	O
it	O
is	O
expected	O
that	O
on	O
average	O
1	O
CPU	B-Device
will	O
always	O
be	O
idle	O
;	O
with	O
21	O
CPUs	B-Device
,	O
2.8	O
will	O
be	O
idle	O
;	O
with	O
40	O
CPUs	B-Device
,	O
19	O
will	O
be	O
idle	O
;	O
with	O
41	O
CPUs	B-Device
,	O
20	O
will	O
be	O
idle	O
.	O
</s>
<s>
Therefore	O
,	O
adding	O
more	O
than	O
40	O
CPUs	B-Device
to	O
that	O
system	O
would	O
be	O
useless	O
.	O
</s>
<s>
In	O
general	O
,	O
for	O
each	O
L/E	O
value	O
,	O
there	O
's	O
a	O
threshold	O
for	O
the	O
maximum	O
number	O
of	O
useful	O
CPUs	B-Device
.	O
</s>
<s>
To	O
reduce	O
the	O
performance	O
degradation	O
of	O
software	B-Operating_System
lockout	I-Operating_System
to	O
reasonable	O
levels	O
(	O
L/E	O
between	O
0.05	O
and	O
0.1	O
)	O
,	O
the	O
kernel	B-Operating_System
and/or	O
the	O
operating	O
system	O
must	O
be	O
designed	O
accordingly	O
.	O
</s>
<s>
Conceptually	O
,	O
the	O
most	O
valid	O
solution	O
is	O
to	O
decompose	O
each	O
kernel	B-Operating_System
data	B-General_Concept
structure	I-General_Concept
in	O
smaller	O
independent	O
substructures	O
,	O
having	O
each	O
a	O
shorter	O
elaboration	O
time	O
.	O
</s>
<s>
This	O
allows	O
more	O
than	O
one	O
CPU	B-Device
to	O
access	O
the	O
original	O
data	B-General_Concept
structure	I-General_Concept
.	O
</s>
<s>
Many	O
uniprocessor	B-Operating_System
systems	I-Operating_System
with	O
hierarchical	B-Operating_System
protection	I-Operating_System
domains	I-Operating_System
have	O
been	O
estimated	O
to	O
spend	O
up	O
to	O
50%	O
of	O
the	O
time	O
performing	O
"	O
supervisor	O
mode	O
"	O
operations	O
.	O
</s>
<s>
If	O
such	O
systems	O
were	O
adapted	O
for	O
multiprocessing	B-Operating_System
by	O
setting	O
a	O
lock	O
at	O
any	O
access	O
to	O
"	O
supervisor	O
state	O
"	O
,	O
L/E	O
would	O
easily	O
be	O
greater	O
than	O
1	O
,	O
resulting	O
in	O
a	O
system	O
with	O
the	O
same	O
throughput	O
as	O
the	O
uniprocessor	B-Operating_System
despite	O
the	O
number	O
of	O
CPUs	B-Device
.	O
</s>
