Upload
jewel-blair
View
215
Download
0
Tags:
Embed Size (px)
Citation preview
Resource Allocation in Network Virtualization
Jie Wu
Computer and Information Sciences
Temple University
Road Map
1. Motivation and Applications
2. Tracing Back: Embedding
3. Basic Models
4. Extensions
1. Hose model
2. Virtual backbone
5. Looking Forward: Other Fields
6. Conclusions
1. Motivation
Network virtualization (Peterson, Shenker, and
Turner’04)
A number of virtual networks (VNs) co-exist over the
same physical network (PN) (substrate network)
VN: a group of nodes that are connected, with
bandwidth reserved in the underlying network
Implementation: RSVP and MPLS A
CD E
F
B
Applications Coexistence
Flexibility
Manageability
Scalability
Isolation
Heterogeneity
ISP = SP + InP
SP: Service Provider
InP: Infrastructure Provider
SDN Programmable switches and
routers than (using virtualization) can process packets for multiple isolated networks
Virtualization Data center networks (DCNs)
2. Tracing Back: Embedding
Embedding (E) of tasks (G) in processors (G’)
Dilation of an edge of G is the length of the path in G’
onto which an edge of G is mapped. Dilation of E is the
maximum edge dilation of G.
Expansion of G is the ratio of the number of nodes in G to
the number of nodes in G’.
Congestion of E is the maximum number of paths
containing an edge in G’, where every path represents an
edge in G.
Load of an E is the maximum number of tasks of G
assigned to any processor of G’.
Embedding Examples
Virtualization Examples
(a) optimal link stress (b) optimal node stress
3. Basic Models
Embed VNs in PN
Subject to CPU (node)
and bandwidth (link)
constraints
General VN embedding NP-hard (multiway
separator problem)
Special VN embedding (fixed nodes) Multicommodity flow
problem
An example of VN embedding
A
C D E
F
B
d e
b a
c
10 10
5
5 5
70
60
90
40
80
40
10
10
2040
20
40 40
40
15 25
25
d e
c
b
a
2010
10
Minimum Cost Multicommodity Flow Multicommodity flow
Capacity constraints, flow
conservation, demand satisfaction
Minimum cost
Sum of a(u, v) f(u, v) on edge (u, v)
Integer flow: hard
Fractional flows: solvable
(Yu et al 06)
Path split
Path migration
Illustration of the benefit of path splitting
Illustration of the benefit of path migration
A
C D
B
20
20
20
20
1040
30
20
10a
a b5 5
20
b
20
20
Existing VN
request 1
A
C D
B
20
20
20
20
1040
30
20
10a
bcd
20
20
New VN request 2 c d
10 1020
10 10
10
a b5 5
20Existing
VN request 1
A
C D
B
20
20
20
20
1040
30
20
10a
bcd
10
10
New VN request 2 c d
10 1030
10 10
20
10
10
A
C D
B
20
20
20
20
1040
30
20
10a
bcd
20
20
Scheduling of Network Updates
Dionysus (Jin et al’14)
Loop freedom
Congestion freedom
Special constraint
A link must occur after an update
that removes an existing flow
Dynamic scheduling
Dependency graph
(Resource allocation graphs)
A network update example. Flows are labeled with their sizes
(a) Current State (b) Target State
A
C D
B
20
20
20
20
2020
20
20
20
F3: 10
F2: 10
A
C D
B
20
20
20
20
2020
20
20
20
F2: 10
F3: 10
F1: 10
F1: 10
One sequence: [F1ĺ F3][F2]
Other sequence: [F1] [F2ĺ F3]
Scheduling of Network Updates
Schedulability
Extension
Introducing intermediate
steps
A Deadlock Example
A
C D
B
20
20
20
20
2020
20
20
20
F3: 20
(a) Current State (b) Target State
F2: 20
A
C D
B
20
20
20
20
2020
20
20
20
F2: 20
F3: 20
(a) Current State (d) Target State
A
C D
B
20
20
20
20
2020
20
20
20
F3: 20
(b) Middle State 1 (c) Middle State 2
A
C D
B
20
20
20
20
2020
20
20
20
F2: 20
F3: 20
F2: 20
A
C D
B
20
20
20
20
2020
20
20
20
F3: 20
F2: 20
A
C D
B
20
20
20
20
2020
20
20
20
F2: 20
F3: 20
4. Extensions: Hose Model (Duffield, Goyal, and Greenberg’99)
Hose: aggregate traffic to
and from endpoints in a VN
Routing structures
Pipe
Ingree (Egree) tree
Shared tree
Mesh
E.g. X (in 3), Y (out 2), and Z
(out 2) using a Steiner tree
Extensions: Virtual Backbone
Mapping VNs onto a shared
substrate (Lu and Turner’06)
Backbone-star, a complete
graph, a ring or a star
Connected dominating set
(CDS) (Wu and Li’99)
A subset (V) of nodes such that
all other nodes not in V have
at least one neighbor in V
Resilience (Dai and Wu’05) K-covered CDS: each node
has k CDS nodes in its 1-hop neighborhood (including itself)
K-connected CDS: can tolerate k-1 faults and still connected
Challenges
Different modelsStaticDynamic (long-term statistical guarantees)
QoS Different provisioning models
Different measurements Minimization of weighted
sum of maximum values of node and link stress
Minimization of long term average value of the weighted sum of bandwidth and CPU revenue
QoS-based Slice Provisioning
Safe vs. Unsafe In terms of available
network resource QoS-based slice provisioning
Slice reservation in unsafe areas
Other extensions K-hop CDS: A subset V such
that each node not in V can reach a node in V within k hops
K-spanner: A spanning subgraph S in which every two vertices are at most k times as far apart in S than on G
6. Looking Forward: Other Fields Virtualization in data
center networks
Virtual machines (VMs)
assignment in physical
machines (PMs)
Subject to CPU and
network bandwidth
constraints
Virtualization in DSN Hadoop scheduling: map,
shuffle, and reduce
Split 1
Split 2
Split 3
Split N
Input data
Mapper 1
Mapper 2
Mapper 3
Mapper N
Map Phase
Reducer 1
Reducer 2
Reducer N
Reduce Phase
Shuffle
Virtualization in SDNs
Virtualization of
controller in SDNs
Multiple controllers
Disjointed
Overlapped (token-
based access control)
Controller placement
Hose Model in DCNs
Elasticity (Li, Wu, and Blaisse’12)
The CPU / bandwidth utilization is the ratio of
the used CPU / bandwidth among all PMs / links
The combined utilization is the maximal one of
the CPU and bandwidth utilizations (bottleneck)
Minimizing the combined
utilization To provide flexibilities for new VM requests
(elasticity)
Hose Model in DCNs (cont’d)
Iterative stack up
Layer by layer
recursive placement
CPU bottleneck: load
balancing placement
Link bottleneck: load
unbalancing
placement
Conclusions
Allocation
centralized vs. distributed
Reconfiguration
migration and dynamic scheduling
Survivability and Flexibility
resource overprovisioning and controlled slicing
Other Applications SDNs and DCNs
Future Challenges
Performance guarantee
Deterministic vs. statistic
Resource discovery and allocation
Cooperation and competition between IPs
Heterogeneity and diversity of infrastructure