C (Stars, M. C. Escher) onsiglio N Istituto di C · lte prestazioni (CERE-CNR, CEntro di studio...
Transcript of C (Stars, M. C. Escher) onsiglio N Istituto di C · lte prestazioni (CERE-CNR, CEntro di studio...
A M
anagement A
rchitecturefor A
ctive Netw
orks
A. Barone, P. Chirco, G
. Di Fatta, G
. Lo Re
AM
S2002, Active M
iddleware Service
Edinburgh (UK
), 23 July 2002(Stars, M
. C. Escher)
ICA
R-C
NR
Istituto di Calcolo e Reti ad Alte prestazioni
(CERE-CNR, CEntro di studio sulle Reti di Elaboratori)
Consiglio Nazionale delle Ricerche (Palerm
o, Italy)
G. D
i Fatta, Computer N
etworks G
roup, ICAR-CN
R2
Overview
•N
etwork M
anagement
–SN
MP fram
ework
–A
ctive Netw
orks for Netw
ork Managem
ent
•A
ctive Netw
orks Managem
ent–
Active M
IB and Active Local A
gent–
Active Routers Testbed and A
Ngate
•M
anager and Monitor tools
•N
etwork Events M
ining•
Conclusions
G. D
i Fatta, Computer N
etworks G
roup, ICAR-CN
R3
Netw
ork Managem
ent•
Managing Entity
•M
anaged Device
–Local A
gent–
MIB (M
anagement
Information Base)
•SN
MP (Sim
ple Netw
orkM
anagement Protocol)
Managing E
ntity
Managed D
eviceM
anaged Device
Managed D
eviceM
anaged Device
Netw
orkM
anagement
Protocol
¸centralized approach: polling¸
active local agent role: simple traps
¸issues: scalability and efficiency
G. D
i Fatta, Computer N
etworks G
roup, ICAR-CN
R4
Active N
etworks for N
etwork M
anagement
Active N
etworks benefits for N
etwork M
anagement are:
• availability of information in interm
ediate nodes
• data processing along the path
• distributed, cooperative, autonomous strategies
(e.g., SmartPackets)
Active N
etworks M
anagement
AN
allow and prom
ote the fast and dynamic introduction of new
servicesand applications. A
ctive Applications are difficult to be debugged and
monitored.
Active N
etworks require a m
ore efficient approach for network and
applications managem
ent.
G. D
i Fatta, Computer N
etworks G
roup, ICAR-CN
R5
Active Routers Testbed (A
RT)
Aim
s of the AR
T project:•
testing new netw
ork services and protocols•
developing new active applications
•netw
ork resource optimisation
40 Active N
odes• Static Topology Configuration (80 peer-to-peer links)
• Dynam
ic Topology Configuration (240port ethernet sw
itch with m
anagement)
A 40 nodes cluster at ICA
R-CNR to study and test A
ctive Netw
orks
AR
T is a hardware platform
which prototypes a
network com
putational environment according to
the Active N
etworks paradigm
.
Practical need for an Active
Managem
ent Architecture
G. D
i Fatta, Computer N
etworks G
roup, ICAR-CN
R6
AN
Managem
ent Framew
orkM
anagement A
pplication Interface
GU
I
XM
Lw
rapper
request[X
ML
]
response [XM
L]
Active N
etwork
EE
EE
EE
EE
EE
ALA
ALA
ALA
ALA
ALA
The Active Local A
gent ( ) is anactive
service of
the EE
which
provides the managem
ent API.
ALA
AN
Access P
oint
XM
L parser
EE
interface
active packets[M
L, Java, …
]
reports
AN
gateway
G. D
i Fatta, Computer N
etworks G
roup, ICAR-CN
R7
Active Local A
gent ( )
AM
IB Objects are variables of A
ctive Applications. A
As can explicitly
submit data to A
LA and associate Filters w
ith data.
A Filter is a test executed w
hen the data value changes. An Event is raised if
the test succeeds.
An A
ction is a capsule associated with an Event and it is released w
hen thatEvent is raised. Events have a local scope and a default A
ction is defined tosim
ply report the Event.
<Filter1, Event1>...<Filter-k, Event-k>
Dataset_data
new value
raise local events
ALA
is a SNM
P-like managem
ent agent which exploits A
N advantages: it
provides the
API
for an
object-oriented A
ctive M
IB and
it is
programm
able according to a Filter-Event-Action m
odel.
AM
IB Object
execute actions
ALA
G. D
i Fatta, Computer N
etworks G
roup, ICAR-CN
R8
ALA
Architecture
ActiveMIB
Get/Setdata, filters,actions
ActionsScheduler
SyncEvents
Generator
Easync
Esync
ActionPerformer
NTP
ActivePacket
<Event,Action>
AgentInterfaceand PolicyEnforcement
ALA - Active Local Agent
Action result:- ActivePacket- RemoteReport- Local/RemoteSet/Get data,filters,actions
Clock andsync process
G. D
i Fatta, Computer N
etworks G
roup, ICAR-CN
R9
AN
gatewayExam
plesIm
plementation
Services
setVarV
alue, getVarV
alue,getA
ppList, setFilterTest,… getTopology, ping,delivery_Path,delivery_Tree, …
setTopology,bootstrap_A
N,
shutdown_A
N, …
ALA
as PLAN
service(O
Caml)
AA
-leveladm
inistrativeand public
packet-level (PLAN
)A
N-level
public
UN
IX Shell scripts
AN
-leveladm
inistrative
G. D
i Fatta, Computer N
etworks G
roup, ICAR-CN
R10
AN
gate - the Active N
etwork gatew
ay
• gateway server selection
• EE selection (only PLAN
is supported in v1.0b6)
• user authentication (user, administrator)
• AN
Manager m
odule
• AA
Monitor m
odule
http://angate.cere.pa.cnr.it
username: guest
password: guest
G. D
i Fatta, Computer N
etworks G
roup, ICAR-CN
R11
Active N
etworks M
anager
• Topology editor
• PLAN
et configuration and setup (only admin)
• Topology discovery
• Link and node status
• Capsule injection with
navigation patterns¸
node¸
path¸
star¸
tree
G. D
i Fatta, Computer N
etworks G
roup, ICAR-CN
R12
Navigation Pattern: D
elivery_MST
23
14
05 6
78
9
Forw
ard Path
Backw
ard Path
injection node
G. D
i Fatta, Computer N
etworks G
roup, ICAR-CN
R13
Active A
pplications Monitor
• ALA
services GU
I
• AN
ode/AA
/AM
IB-data tree
• AA
tracing and debugging and perform
ance measurem
ents
• Message area for reports
G. D
i Fatta, Computer N
etworks G
roup, ICAR-CN
R14
New
Applications
Managem
entA
pplication
GU
I
XM
Lw
rapper
AN
Access P
oint
Active N
etwork
XM
L parser
EE
Interface
active packets[M
L, Java, …
]reports
EE
EE
EE E
E
EE
AN
gateway
Managem
entA
pplication
GU
I
XM
Lw
rapper
Netw
ork EventsM
iningL
ogicE
ngineX
ML
wrapper
G. D
i Fatta, Computer N
etworks G
roup, ICAR-CN
R15
Netw
ork Events Mining (N
EM)
NEM
systems deals w
ith large archives of events gathered fromthe netw
ork devices to extract useful information.
Typical NEM
goals are:• the diagnose of root causes of netw
ork faults and performance
degradations by establishing relationships between netw
ork events;• filtering event (alarm
) flood by correlating several events into asingle conceptual event.
NEM
systems should provide correctness and optim
ality.
G. D
i Fatta, Computer N
etworks G
roup, ICAR-CN
R16
NEM
Application
The Logical Managing Entity is based on the Situation C
alculus,w
hich is a dialect of the First Order Logic and allow
s them
odeling of dynamic system
s. It is implem
ented by the Golog
logic programm
ing language.
Netw
ork Events Mining
Logic
Engine
XM
Lw
rapper
XM
L request/responseto/from
AN
Situation Calculus• entities• events• actions• situations• predicates• fluents
G. D
i Fatta, Computer N
etworks G
roup, ICAR-CN
R17
The Logical System D
esign
1. O
ntological engineering process providesß
effective structured representation of all the network
entities, events and actions.(fi
predicates, fluents, actions, preconditions)
2. Logical axiom
atization of relationships among the
network events are to be defined.
ßThe occurrence of an action a m
akes the system m
ovefrom
a given situation si to the consequent situation si+1.(fi
successor state axioms)
G. D
i Fatta, Computer N
etworks G
roup, ICAR-CN
R18
Ontological A
nalysis
Logical entities and relationships are expressed as Predicatesand Fluents.
Predicates:ß iface(I)
ß address(I, <string>)
ß node(N)
ß flag_node(N)
ß ala(A)
ß ala_location(A, N)
ß link(L)
ß connect(L, I1, I2) where link(L), iface(I1), iface(I2)
ß belongs(I, N)
· · ·
connect(L, I1, I2)
where link(L),
iface(I1), iface(I2)
G. D
i Fatta, Computer N
etworks G
roup, ICAR-CN
R19
Fluents
In situation s the flag of node N is on\off.
node_flag_status(N, on\off, s)
In situation s node N is on\off.
node_status(N, on\off,
s)
In situation s the Active Local Agent A is
on\off.
ala_status(A, on\off,
s)
In situation s interface I is on\off.
iface_status(I, on\off,
s)
In situation s nodes A and B are connected at
network layer by path L.
path(A, B, L,
s)
In situation s nodes N1 and N2 are directly
connected at network layer.
neighbour(N1, N2,
s)
In situation s <arrayTable> is the routing
table of node N.
routing_table(N,<arrayTable>,s)
In situation s the physical link L is on\off.
link_status(L, on\off,
s)In situation s nodes A
and B are
connected at network layer by
path L.path(A, B,
L, s)
G. D
i Fatta, Computer N
etworks G
roup, ICAR-CN
R20
Primitive A
ctions
...
...
The semaphore of the routing table in the router R is
switched off.
Turn_flag_off(R)
Router R forwards a packet P according to its current
routing table T.
Forward(R, P, T)
Router R executes a Shortest Path Tree algorithm to
determine the shortest paths table T toward all the
destinations.
SPT_execute(R, T)
Router R stores a routing table change (this will allow
to trace back the events).
Update_store(R, entry)
Router R receives a topology change from a neighbour
(the semaphore of the routing table is switched on).
Update_receive(R, upd)
Router R sends a topology change to all its neighbours.
Update_send(R, upd)
The network link L is set down (the semaphore of both
the connected routers is switched on).
Link_down(L)
The network link L is set up (the semaphore is switched
on in both the connected routers).
Link_up(L)
Router R stores a routing table change(this allows to trace back events).
Update_store(R, entry)
G. D
i Fatta, Computer N
etworks G
roup, ICAR-CN
R21
Primitive A
ction Preconditionsß Poss
(Link_up(L), s)
≡ link_status(L, off, s).
ß Poss(Link_down(L), s)
≡ link_status(L, on, s).
ß Poss(Update_send(R, upd), s)
≡ node_flag_status(R, on, s).
ß Poss(SPT_execute(R, T), s)
≡ node_flag_status(R, on, s).
ß Poss(Forward(R, PCK, T), s)
≡ Destination(PCK) = D
Ÿ Routing_entry(T,D,s) = I
Ÿ iface_status(I,on,s).
ß Poss(Turn_flag_off(R), s)
≡ node_flag_status(R, on, s).
· · ·
Successor State Axiom
sß neighbour(N1,N2,do
(a,s))
≡ neighbour(N1,N2,s)
Ÿ connect(L,I1,I2)
Ÿ belongs(I1,N1)
Ÿ belongs(I2,N2)
Ÿ a
≠ Link_down(L).
ß node_flag_status(N, on, do
(a, s))
≡ node_flag_status(N, on, s)
Ÿ a
≠ Turn_flag_off(N).
· · ·
G. D
i Fatta, Computer N
etworks G
roup, ICAR-CN
R22
NEM
goals
:- path(node1, node2, X, s1).
yes X=[node1, node5, ..., node2]
Finally, we can express goals of the logical engine such
as:
:- path(node1, node2, X, s2).
no
or
G. D
i Fatta, Computer N
etworks G
roup, ICAR-CN
R23
Conclusions and ongoing work
The A
N
Managem
ent fram
ework
exploits netw
orkprogram
mability to enable agent-based distributed strategies.
Main features are:
• extension of the traditional MIB objects into m
orepow
erful objects with user custom
izable code.
• agent programm
ability according to the Filter-Even-A
ction model
New
m
anagement
applications can
be devised
to take
advantage of the distributed and autonomous agent control.
A N
etwork Events M
ining application is under development.