<s>
Temporal	B-General_Concept
isolation	I-General_Concept
or	O
performance	O
isolation	O
among	O
virtual	B-Architecture
machine	I-Architecture
(	O
VMs	O
)	O
refers	O
to	O
the	O
capability	O
of	O
isolating	B-General_Concept
the	I-General_Concept
temporal	I-General_Concept
behavior	I-General_Concept
(	O
or	O
limiting	O
the	O
temporal	O
interferences	O
)	O
of	O
multiple	O
VMs	O
among	O
each	O
other	O
,	O
despite	O
them	O
running	O
on	O
the	O
same	O
physical	O
host	O
and	O
sharing	O
a	O
set	O
of	O
physical	O
resources	O
such	O
as	O
processors	O
,	O
memory	O
,	O
and	O
disks	B-Device
.	O
</s>
<s>
One	O
of	O
the	O
key	O
advantages	O
of	O
using	O
virtualization	B-General_Concept
in	O
server	O
consolidation	O
,	O
is	O
the	O
possibility	O
to	O
seamlessly	O
"	O
pack	O
"	O
multiple	O
under-utilized	O
systems	O
into	O
a	O
single	O
physical	O
host	O
,	O
thus	O
achieving	O
a	O
better	O
overall	O
utilization	O
of	O
the	O
available	O
hardware	O
resources	O
.	O
</s>
<s>
In	O
fact	O
,	O
an	O
entire	O
Operating	B-General_Concept
System	I-General_Concept
(	O
OS	O
)	O
,	O
along	O
with	O
the	O
applications	O
running	O
within	O
,	O
can	O
be	O
run	O
in	O
a	O
virtual	B-Architecture
machine	I-Architecture
(	O
VM	O
)	O
.	O
</s>
<s>
However	O
,	O
when	O
multiple	O
VMs	O
concurrently	O
run	O
on	O
the	O
same	O
physical	O
host	O
,	O
they	O
share	O
the	O
available	O
physical	O
resources	O
,	O
including	O
CPU(s )	O
,	O
network	O
adapter(s )	O
,	O
disk(s )	O
and	O
memory	O
.	O
</s>
<s>
In	O
a	O
world	O
of	O
computing	O
that	O
is	O
shifting	O
towards	O
cloud	B-Architecture
computing	I-Architecture
paradigms	O
where	O
resources	O
(	O
computing	O
,	O
storage	O
,	O
networking	O
)	O
may	O
be	O
remotely	O
rented	O
in	O
virtualized	O
form	O
under	O
precise	O
service-level	O
agreements	O
,	O
it	O
would	O
be	O
highly	O
desirable	O
that	O
the	O
performance	O
of	O
the	O
virtualized	O
resources	O
be	O
as	O
stable	O
and	O
predictable	O
as	O
possible	O
.	O
</s>
<s>
They	O
aim	O
to	O
achieve	O
some	O
degree	O
of	O
temporal	B-General_Concept
isolation	I-General_Concept
across	O
the	O
concurrently	O
running	O
VMs	O
,	O
at	O
the	O
various	O
critical	O
levels	O
of	O
scheduling	O
:	O
CPU	B-Device
scheduling	O
,	O
network	O
scheduling	O
and	O
disk	B-Device
scheduling	O
.	O
</s>
<s>
For	O
the	O
CPU	B-Device
,	O
it	O
is	O
possible	O
to	O
use	O
proper	O
scheduling	O
techniques	O
at	O
the	O
hypervisor	O
level	O
to	O
contain	O
the	O
amount	O
of	O
computing	O
each	O
VM	O
may	O
impose	O
on	O
a	O
shared	O
physical	O
CPU	B-Device
or	O
core	O
.	O
</s>
<s>
For	O
example	O
,	O
on	O
the	O
Xen	B-Operating_System
hypervisor	I-Operating_System
,	O
the	O
BVT	O
,	O
Credit-based	O
and	O
S-EDF	O
schedulers	O
have	O
been	O
proposed	O
for	O
controlling	O
how	O
the	O
computing	O
power	O
is	O
distributed	O
among	O
competing	O
VMs	O
.	O
</s>
<s>
To	O
get	O
stable	O
performance	O
in	O
virtualized	O
applications	O
,	O
it	O
is	O
necessary	O
to	O
use	O
scheduler	O
configurations	O
that	O
are	O
not	O
work-conserving	B-General_Concept
.	O
</s>
<s>
Finally	O
,	O
with	O
a	O
multi-core	B-Architecture
or	O
multi-processor	B-Operating_System
physical	O
host	O
,	O
it	O
is	O
possible	O
to	O
deploy	O
each	O
VM	O
on	O
a	O
separate	O
processor	O
or	O
core	O
to	O
temporally	O
isolate	O
the	O
performance	O
of	O
various	O
VMs	O
.	O
</s>
<s>
Also	O
,	O
it	O
is	O
possible	O
to	O
install	O
multiple	O
network	B-Protocol
adapters	I-Protocol
on	O
the	O
same	O
physical	O
host	O
,	O
and	O
configure	O
the	O
virtualization	B-General_Concept
layer	O
so	O
that	O
each	O
VM	O
may	O
grant	O
exclusive	O
access	O
to	O
each	O
one	O
of	O
them	O
.	O
</s>
<s>
For	O
example	O
,	O
this	O
is	O
possible	O
with	O
the	O
driver	O
domains	O
of	O
the	O
Xen	B-Operating_System
hypervisor	I-Operating_System
.	O
</s>
<s>
Multi-queue	O
network	B-Protocol
adapters	I-Protocol
exist	O
which	O
support	O
multiple	O
VMs	O
at	O
the	O
hardware	O
level	O
,	O
having	O
separate	O
packet	O
queues	O
associated	O
to	O
the	O
different	O
hosted	O
VMs	O
(	O
by	O
means	O
of	O
the	O
IP	O
addresses	O
of	O
the	O
VMs	O
)	O
,	O
such	O
as	O
the	O
Virtual	B-Architecture
Machine	I-Architecture
Device	O
Queue	O
(	O
VMDq	O
)	O
devices	O
by	O
Intel	O
.	O
</s>
<s>
Finally	O
,	O
real-time	O
scheduling	O
of	O
the	O
CPU	B-Device
may	O
also	O
be	O
used	O
for	O
enhancing	O
temporal	B-General_Concept
isolation	I-General_Concept
of	O
network	O
traffic	O
from	O
multiple	O
VMs	O
deployed	O
on	O
the	O
same	O
CPU	B-Device
.	O
</s>
<s>
When	O
using	O
real-time	O
scheduling	O
for	O
controlling	O
the	O
amount	O
of	O
CPU	B-Device
resources	O
reserved	O
for	O
each	O
VM	O
,	O
one	O
challenging	O
problem	O
is	O
properly	O
accounting	O
for	O
the	O
CPU	B-Device
time	O
applicable	O
to	O
system-wide	O
activities	O
.	O
</s>
<s>
For	O
example	O
,	O
in	O
the	O
case	O
of	O
the	O
Xen	B-Operating_System
scheduler	O
,	O
the	O
Dom0	B-Operating_System
and	O
the	O
driver	O
domains	O
services	O
might	O
be	O
shared	O
across	O
multiple	O
VMs	O
accessing	O
them	O
.	O
</s>
<s>
Similarly	O
,	O
in	O
the	O
case	O
of	O
the	O
KVM	B-Application
hypervisor	O
,	O
the	O
workload	O
imposed	O
on	O
the	O
host	O
OS	O
due	O
to	O
serving	O
network	O
traffic	O
for	O
each	O
individual	O
guest	O
OS	O
might	O
not	O
be	O
easily	O
distinguishable	O
,	O
because	O
it	O
mainly	O
involves	O
kernel-level	O
device	O
drivers	O
and	O
the	O
networking	O
infrastructure	O
(	O
on	O
the	O
host	O
OS	O
)	O
.	O
</s>
<s>
Some	O
techniques	O
for	O
mitigating	O
such	O
problems	O
have	O
been	O
proposed	O
for	O
the	O
Xen	B-Operating_System
case	O
.	O
</s>
<s>
Along	O
the	O
lines	O
of	O
adaptive	O
reservations	O
,	O
it	O
is	O
possible	O
to	O
apply	O
feedback-control	O
strategies	O
to	O
dynamically	O
adapt	O
the	O
amount	O
of	O
resources	O
reserved	O
to	O
each	O
virtual	B-Architecture
machine	I-Architecture
to	O
maintain	O
stable	O
performance	O
for	O
the	O
virtualized	O
application(s )	O
.	O
</s>
<s>
Following	O
the	O
trend	O
of	O
adaptiveness	O
,	O
in	O
those	O
cases	O
in	O
which	O
a	O
virtualized	O
system	O
is	O
not	O
fulfilling	O
the	O
expected	O
performance	O
levels	O
(	O
either	O
due	O
to	O
unforeseen	O
interferences	O
of	O
other	O
concurrently	O
running	O
VMs	O
,	O
or	O
due	O
to	O
a	O
bad	O
deployment	O
strategy	O
that	O
simply	O
picked	O
up	O
a	O
machine	O
with	O
insufficient	O
hardware	O
resources	O
)	O
,	O
it	O
is	O
possible	O
to	O
live-migrate	O
virtual	B-Architecture
machines	I-Architecture
while	O
they	O
are	O
running	O
,	O
so	O
as	O
to	O
host	O
them	O
on	O
a	O
more	O
capable	O
(	O
or	O
less	O
loaded	O
)	O
physical	O
host	O
.	O
</s>
