NGN statement on e2e network performance froze in 2008 with the Y.2173 (former Y.mnm) standard which specified a method for coordination across domains, involving active probing, parsing and sharing results, etc. Today, it is widely accepted that future network functions is to be virtualized and controlled by software. However, virtual switches -- the best models of which to the day are based on MiniOS/ClickOS/ClickRouter designs -- mimic the blackboxing policy found in clouds. This paper shows that if this policy is revoked in favor of declarative coordination, complexity for optimization of overlaying service networks can be substantially reduced. The proposed case is represented by domains declaring their resources/capabilities (e.g. e2e throughput and delay) openly, thus allowing service networks to coordinate among each other. Traditional approach is represented by the VNE optimization problem.
2. .
This Talk is about Cloud Networks
• a network of DCs is a cloud -- say, Amazon EC2 with 9 regions
• a network of clouds is a federation -- say, Amazon x 1dc + IDCF x
20dcs in Japan
◦ we already start hitting the heterogeneity wall, since there is no solution yet, no
such federations exist today, but some are planned 03
• .. moving further to network edge, we find Akamai with its 30k+ cloudlets
02 -- a class of its own
◦ heterogeneity problem? in hardware NO, but in e2e network performance STRONG YES
◦ but, Akamai has all the control, so, they manage to balance the load properly
• ... now, having incorporated all the devices at network edge you have the
fog cloud
• ... on a side note... why cloud networks? ... well, because most networking
soon will be done inside and/or by clouds
03 "Chameleon project" https://www.chameleoncloud.org (2014)
02 B.Frank+8 "Pushing CDN-ISP Collaboration to the Limit" ACM SIGCOMM Computer Communication Review (2013)
M.Zhanikeev -- maratishe@gmail.com The Declarative-Coordinated Model for Self-Optimization of Service Networks -- bit.do/160121 2/17
...
2/17
3. .
Cloudification Timeline and Scope
WiFi
Scale/sizeofcloudifieditems
Timeline
Mid-2015
Probably
should not
cloudify
Should
definitely
cloudify
Data
Centers
Racks IoV > VC
Desktop
APs,
Beacons
Smartphone
Sensor
IoT
Delegated
networking
M.Zhanikeev -- maratishe@gmail.com The Declarative-Coordinated Model for Self-Optimization of Service Networks -- bit.do/160121 3/17
...
3/17
4. .
One Cloud Service
.
VNE Problem
..
....stands for Virtual Network Embedding
APP
Cloud/DC
APP
APP
VM
Container
APP
Cloud/DC
APP
APP
….
• service can provide the
graph of its apps
• graph can include end users in
form of DC-user e2e
network performance
• federated cloud known its
actual topology and can use
VNE to map 04
04 M.Zhanikeev "A New VNE Method for More Responsive Networking in Many-to-Many Groups" 7th ICUFN (2015)
M.Zhanikeev -- maratishe@gmail.com The Declarative-Coordinated Model for Self-Optimization of Service Networks -- bit.do/160121 4/17
...
4/17
5. .
OSPF, QoS Routing → VNE
• OSPF optimizes one parameter (cost/weight) for one physical graph, and is NP
Hard
• QoS Routing optimizes multiple parameters for one path, and is NP Hard
• VNE maps one virtual graph on top of a physical graph and is NP Hard-er?
◦ both OSPF and QoS Routing generalize into a VNE problem if complexity is ignored
M.Zhanikeev -- maratishe@gmail.com The Declarative-Coordinated Model for Self-Optimization of Service Networks -- bit.do/160121 5/17
...
5/17
6. .
VNE: Virtual Network Embedding
.
VNE is all about...
..
.
... multiple virtual graphs mapped onto
one real/physical topology/graph
Physical
layer
Virtual
Layer1
VN
Request
2
Make VN
3 VN
Reply
4 Virtual
Virtual
Physical
1. service sends the request in form of a
virtual graph
2. NOC casts its magic 04 -- here we
depend on a good VNE heuristic (so far,
non-existent)
3. the mapping (IPs, ports, etc.) is
returned back to the service or,
otherwise, the request is denied
4. the service uses its new service
network
04 M.Zhanikeev "A New VNE Method for More Responsive Networking in Many-to-Many Groups" 7th ICUFN (2015)
M.Zhanikeev -- maratishe@gmail.com The Declarative-Coordinated Model for Self-Optimization of Service Networks -- bit.do/160121 6/17
...
6/17
7. .
The Half-Way Point
M.Zhanikeev -- maratishe@gmail.com The Declarative-Coordinated Model for Self-Optimization of Service Networks -- bit.do/160121 7/17
...
7/17
8. .
Solution to the Complexity Problem
.
The Obvious Solution is...
..
.... the DiY design, that is, to let services self-optimize
• however....
◦ it is one thing to self optimize in a non-cooperative environment
◦ ... and quite another when clouds offer tools/envs/platforms that
help self-optimize better
.
Cloud Probing Technology...
..
....is the best choice for non-cooperative environments
M.Zhanikeev -- maratishe@gmail.com The Declarative-Coordinated Model for Self-Optimization of Service Networks -- bit.do/160121 8/17
...
8/17
9. .
DiY (1) DC Probing
• visualization of active probing
across 8 Amazon regions 05
california
ireland
oregon
saopaulo
singapore
sydney
tokyo virginia
same DC
large bulk transfer
california
ireland
oregon
saopaulo
singapore
sydney
virginia
DC - other DC
e2e delay
05 M.Zhanikeev "Performance Management of Cloud Populations via Cloud Probing" IPSJ JIP (related to http://goo.gl/f3QpiW) (2015)
M.Zhanikeev -- maratishe@gmail.com The Declarative-Coordinated Model for Self-Optimization of Service Networks -- bit.do/160121 9/17
...
9/17
10. .
DiY (2) Heroku Probing
• this time probing between Amazon VMs and heroku dynos
• obvious outcome (figures are vector=zoomable): heroku lives in Virginia DC
• see details in 14
14 M.Zhanikeev "Heroku is Best Modeled as a Bottle" http://goo.gl/j6JzPz (2015)
M.Zhanikeev -- maratishe@gmail.com The Declarative-Coordinated Model for Self-Optimization of Service Networks -- bit.do/160121 10/17
...
10/17
12. .
Declarative-Coordinated Model
.
It is all about...
..
.
...each provider declaring its own
resources autonomously to facilitate
coordination when optimizing service
networks
Network
Provider s
(NOC)
Agent
Service Traditional agents
True Autonomy
• declaration
◦ each NOC creates and declares its own
resources
◦ example: e2e circuits to other
domains?
• autonomy
◦ the point is that declaration can be
done completely autonomously,
no need for global QoS management like
in NGN
• coordination
◦ under autonomy, optimization can
be achieved via coordination, instead of
the traditional VNE with the NP-Hard
level complexity
M.Zhanikeev -- maratishe@gmail.com The Declarative-Coordinated Model for Self-Optimization of Service Networks -- bit.do/160121 12/17
...
12/17
13. .
The New NOC
NOC AgentsService
Local
monitoring
Req /Rep Optimize (VNE)
Discover
S-A Coordination
A-A
Coordination
N-A Coordination
TraditionalProposed
M.Zhanikeev -- maratishe@gmail.com The Declarative-Coordinated Model for Self-Optimization of Service Networks -- bit.do/160121 13/17
...
13/17
14. .
Numeric Analysis
M.Zhanikeev -- maratishe@gmail.com The Declarative-Coordinated Model for Self-Optimization of Service Networks -- bit.do/160121 14/17
...
14/17
15. .
Abilene Network and Setup
.
Abilene Network
..
.
...11 nodes, 27 links, a real network in US, only topology is used, while node/link
capacities are ignored (set to unit measures)
• VNE: implemented a standard method, see
review at 04
• Proposal: each node in a free agent,
the service communicates to each and
selects the one that provides the best
choice at a given time -- the same stress
(but distributed) is used
04 M.Zhanikeev "A New VNE Method for More Responsive Networking in Many-to-Many Groups" 7th ICUFN (2015)
M.Zhanikeev -- maratishe@gmail.com The Declarative-Coordinated Model for Self-Optimization of Service Networks -- bit.do/160121 15/17
...
15/17
16. .
Results
0 2.95 5.9 8.85 11.8 14.75 17.7 20.65 23.6 26.55
Decreasing order
0
20
40
60
80
100
Linkcongestion VNECoordination (proposed)
Link states
0 2 4 6 8 10
Decreasing order
1.68
1.74
1.8
1.86
1.92
1.98
2.04
2.1
2.16
2.22
Nodecongestion
Node states
• link congestion is in %
• node congestion is profile from 04 --
log of sum of link congestions, basically
• one episode -- kept adding
hub-n-spokes virtual networks until
full
• coordination method
drastically outperforms the
traditional VNE
04 M.Zhanikeev "A New VNE Method for More Responsive Networking in Many-to-Many Groups" 7th ICUFN (2015)
M.Zhanikeev -- maratishe@gmail.com The Declarative-Coordinated Model for Self-Optimization of Service Networks -- bit.do/160121 16/17
...
16/17
17. .
That’s all, thank you ...
M.Zhanikeev -- maratishe@gmail.com The Declarative-Coordinated Model for Self-Optimization of Service Networks -- bit.do/160121 17/17
...
17/17