011712 ciena ipanema webinar final
TRANSCRIPT
Beyond MPLS: The Network as the Cloud-
Computing Delivery Platform
A Light Reading Webinar
Sponsored by
Welcome to Today's Light Reading Webinar
Participate in the webinar: Ask questions, share
feedback via the survey, and access the Information
panel.
Personalize your experience: Click the buttons at
the bottom of your screen to open supporting content
and user tools at your own convenience.
Technical Issues: Ask the support team for live
assistance in the ask-a-question window.
Enjoy the webinar and thank you for viewing!
Today’s Presenters
Mitchell H. Auster Senior Advisor
Market Development
Ciena
Sarah Wallace Analyst
Heavy Reading
David White Senior VP Global
Business Development
Ipanema Technologies
About This Webinar
This webinar is intended for Carriers, Network Service Providers
and Network Managers who are considering offering cloud services
as part of their revenue stream. This webinar will discuss:
Why service providers are unique in offering cloud services
What challenges may come in preparing their networks to offer
cloud-based services
How cloud computing increases the complexity of enterprise
WANs
How MPLS is a 15-year old technology and under threat of
revenue erosion from enterprise IT transformations and a
commoditization of MPLS services
solutions?
Why Cloud
The Appeal of Cloud Services
• Reduction of cost
- Operations
- Maintenance
- Paid for incrementally
• Increased Storage
- Organizations can store
more data than on
private computer systems
• More resources and staff
- Agility
- Security
- Privacy
• Scalability
- Multi-tenant environment
- From 1,000 clients to 10,000
very quickly
Amazon
(EC2)
New York Times
FourSquare Hootsuite Quora Reddit SCVNGR
“Cloudpocalypse” … When the cloud fails
Amazon reports trouble due to “excessive re-mirroring of its Elastic Block Storage (EBS) volumes and that a “networking event” caused a domino effect across availability zones in that region, in which many of its storage volumes created new backups of themselves.
That filled up Amazon’s available storage capacity and
prevented some sites from accessing their data.
These sites supported by Amazon‟s EC2 experienced outages, some up to 36 hours.
solutions?
Hence: The Time is Right for Service Providers
INFRASTRUCTURE
STORAGE
Service Providers can provide…
Service Providers are in a prime position to capitalize
on the cloud revolution because they can provide the following
compared to traditional cloud vendors:
MANAGEMENT
VALUE-ADDED VPN SERVICES
The Challenge for Service Providers
Use the existing network infrastructure to offer value
added services
Offer new cloud services using traditional data transport
Meanwhile, delivering all these cloud services in an
environment that is:
Secure
Scalable
In real-time
Reliable
The Key
The key for service providers who want to provide
superior cloud services will be to: Have Superior WAN Optimization and Management
Have Superior Bandwidth Management
Move Beyond Dated MPLS
Beyond MPLS:
Optimizing the Cloud Backbone
www.ciena.com
12 © Ciena Confidential and Proprietary
Key Messages
The ability to place and/or migrate workloads (VMs) in arbitrary data
centers improves resiliency, efficiency and application performance
Before migrating a VM, the associated data store must be reachable at
new DC
Trend to active-active data replication among 2 or more DCs
The ability to migrate a Live VM is becoming table stakes
Minimizing interruption (pause) is critical
The cloud backbone network is a critical determinant of success
Two broad requirements for the inter-DC Cloud Backbone:
Virtualization for scalable multi-tenancy
Orchestrated QoS to optimize profitability
13 © Ciena Confidential and Proprietary
Cloud Service Driven Inter-DC Traffic
Traffic Type Quantity of
Instances
Flow Duration
per Instance
Bandwidth
per Instance
QoS
Sensitivity
Total WAN
Bandwidth
Data Storage Migration Low Short-to-Medium Very High Medium High
Live Virtual Machine Migration Low-High Short-to-Medium Medium High Medium-
Very High
Data Storage
Update/Replication
Medium Medium-to-Long Medium High Very High
Distributed vApp Inter-VM
traffic, VM mgmt (VM-VPN)
Medium Short-to-Long Low Medium Low
Tra
nsactio
nal
perm
an
en
t S
em
i-
High Peak-to-Valley pairwise DC-DC bandwidth demand, especially
driven by high-bw, short-medium duration, QoS-sensitive flows
• Disaster avoidance, DC consolidation or geographic re-location
• ~5000 VMs @ 20 GB 100 TB (~10x to move datastore) Bulk VM Migration
• Workload balancing – capacity, follow sun/moon, power cost
• ~500 VMs @ 20 GB 10 TB (~10x to move datastore) Periodic VM Migration
• Cloudbursting for capacity enhancement • ~50 VMs @ 20 GB 1 TB (~10x to move datastore)
Unplanned VM Migration
14 © Ciena Confidential and Proprietary
0
100
200
300
400
500
600
0 0.5 1 2 3 4 5 6 10 20 50
Latency (ms)
Latency Effect on VMM Completion Time
Completion time increases with network latency
Frame loss rapidly compounds the effect of network latency
Loss %
Loss Effect on VMM Completion Time
(e.g. @ 5 msec RTT) 2500
2000
1500
1000
500
0
0 0.1 0.2 0.3 0.4 0.5 1
VM
M c
om
ple
tio
n T
ime
(s
ec
s)
Bandwidth Effect on VMM Pause Time
Client-Server Business App
Development Workload
Retail Web Server w/ 600 Users
VM
M C
om
ple
tio
n T
ime
(s
ec
s)
VM Migration Performance Effects of Latency, Loss, Bandwidth
Cloud backbone must deliver high bandwidth,
and minimize loss & latency
100
15 © Ciena Confidential and Proprietary
General purpose IP/MPLS Core (typical 99.95% PDR)
ill-suited for certain Inter-DC traffic
High bandwidth, QoS-sensitive flows must be traffic-engineered to ~zero loss
“Real-time traffic like storage replication or live migration requires dedicated
bandwidth, meaning that an optimization solution must be able to prioritize traffic
flows, and prevent them from being pushed aside by other applications.” (Forrester)
TCP/IP flows of
equivalent CoS
Congestion High bandwidth
transactional QoS-
sensitive flows
Low bandwidth,
semi-permanent
QoS-tolerant flows
General Purpose
IP/MPLS Core
TCP is aggressively
egalitarian
each session gets the
same bandwidth 0
500
1000
1500
2000
2500
3000
3500
4000
4500
Packet Delivery Ratio (1 - Loss%)
RTT =1ms
RTT =2ms
RTT =5ms
RTT =10ms
TCP Throughput vs. PDR [RFC 5348]
TC
P t
hro
ug
hp
ut
(Mb
ps)
X
16 © Ciena Confidential and Proprietary
Typical Constrained DC Resource Sharing
X1 = no capacity available, need to burst to provider IaaS cloud service
X2 = no capacity available at local provider DC
X3 = insufficient bandwidth/QoS to migrate live VMs/storage to distant DCs
Compute
Resource capacity
Resource used Storage
Enterprise DC
Compute Storage
Compute Storage
Provider DC 2
(Distant) Compute Storage
User
Provider DC N
(Distant)
Provider DC 1
(Local)
General Purpose IP/MPLS Core
New
workload
X1
X2
X3
X3
Underutilized resources and unmet demand
17 © Ciena Confidential and Proprietary
Cloud Service Provider Inter-DC backbone Elastic, On-Demand, High-Performance Packet-Optical mesh
L1 or L2 Packet Transport Server Connections
L2 S
erv
ice F
low
s
DC-A
DC-B
DC-C
Point-to-Point
Static and On-
Demand
Any-to-Any CBE
CBE
shared
uplink
Common Packet-
Optical Mesh
Converged Data Center Fabric
Virtualized Workloads
VM VM VM
CBE
Extended VM-VPN (E-LAN)
Re-directed User-vApp (E-LAN)
Distributed File System (E-LAN)
VM Migration traffic (E-Line)
Storage Replication (E-Line, OTN)
Storage Migration – (E-Line) Transactional
Static/Semi-permanent
18 © Ciena Confidential and Proprietary
Data Center Without Walls Enables global efficient workload distribution
X1 = no capacity available, need to burst to provider IaaS cloud service
X2 = no capacity available at local provider DC
= performance on-demand to migrate VMs/storage to distant DCs and provide user-to-vApp QoS
Compute Storage
Enterprise DC
Compute Storage
Compute Storage
Provider DC 2
(Distant) Compute Storage
User
Provider DC N
(Distant)
Provider DC 1
(Local)
New
workload
X1
X2
General Purpose IP/MPLS Core
Minimize capacity required; maximize resource utilization & maximize
fulfilled demand Anywhere, anytime provision of resources
Performance On-Demand
Cloud Backbone
Enterprise DC
Compute Storage
User
New
workload
Compute
Storage
Performance On-Demand
Cloud Backbone
Resource capacity
Resource used
19 © Ciena Confidential and Proprietary
Guarantee High Performance to Increase Resource Efficiency
DC
1
DC
2
DC
3
DC
4
DC
5
DC
6
DC
7
DC
9
DC
10
DC
11
Disparate DCs/Resources Resource capacity
Resource used
Virtual DC without Walls –
shared global resource pool
DC
1
DC
2
DC
3
DC
4
DC
5
DC
6
DC
7 DC
8 DC
9
DC
10
DC
11
DC
8
Ensure high BW
and minimize
inter-DC latency
& loss
Increase radius
of candidate
DCs for Live VM
Migration
Increase shared
resource pool
utilization
Minimize
resource
requirement in
each DC
Minimize cost of
total resource
pool required
20 © Ciena Confidential and Proprietary
… but how to address the high peak-to-valley
pairwise DC-DC bandwidth demand?
build full fixed mesh for sum of each
pairwise daily peak?
build stat-mux mesh for daily peak
sum of total forecast demands?
>95% of time
Low: Y Mbps
Cloudburst –
live VM, storage
migration
<5% of time
High: NxY Gbps
En
terp
ris
e Ia
aS
Clo
ud
Pro
vid
er
Fixed connection –
sized for peak
•Wasted bandwidth, $$$
En
terp
ris
e Ia
aS
Clo
ud
Pro
vid
er
Fixed connection –
sized for median
•Insufficient bandwidth
•Low performance, protracted cloudburst
Inter-DC Bandwidth Demand (Enterprise to Provider)
21 © Ciena Confidential and Proprietary
Inter-DC Bandwidth Demand (Provider to Provider)
High pairwise (DCa – DCz) peak-to-valley (& mean) ToD bandwidth variation
DC
DC
DC
DC DC
Window 1 peaks
Sum = X Tbps
DC
DC
DC
DC DC
Window 2 peaks
Sum = 1.5X Tbps
DC
DC
DC
DC DC
Window 3 peaks
Sum = 2X Tbps
DC
DC
DC
DC DC
Window 4 peaks
Sum = 0.25X Tbps
How to
build?
Fixed mesh of
wavelengths/circuits,
sized for pairwise peaks?
DC
DC
DC DC
DC Severely underutilized
bandwidth for large windows
22 © Ciena Confidential and Proprietary
Inter-DC Bandwidth Demand (Provider to Provider)
High pairwise (DCa – DCz) peak-to-valley (& mean) ToD bandwidth variation
DC
DC
DC
DC DC
Window 1 peaks
Sum = X Tbps
DC
DC
DC
DC DC
Window 2 peaks
Sum = 1.5X Tbps
DC
DC
DC
DC DC
Window 3 peaks
Sum = 2X Tbps
DC
DC
DC
DC DC
Window 4 peaks
Sum = 0.25X Tbps
How to
build?
Fixed packet access
wavelengths/circuits, sized
for DCn peak sum + pairwise
stat-mux mesh (e.g. LSPs)?
1) Stat-mux mesh CIRs
sized for pairwise peak
underutilized core
bandwidth, or
2) Stat-mux mesh CIRs
sized for pairwise mean
packet loss on
congested links
DC
DC
DC
DC DC
23 © Ciena Confidential and Proprietary
… how to address the high peak-to-valley pairwise DC-DC
bandwidth demand? Build orchestrated Performance on
Demand for daily peak sum of actual, smoothed demands!
„Network Hypervisor‟
FC
Virtual Machine (VM)
applications
Enterprise Data Center Provider Data Center
Ethernet
FC
Ethernet
FC
SAN
SAN
Ethernet
FC
SAN
Server
SAN
Switch
SAN
Storage
SAN
Server
VPLEX
SAN
Switch
SAN
Storage
VPLEX
Control Plane
Provider “Cloud OS” Billing Systems
Network Hypervisor enables automated performance on demand
24 © Ciena Confidential and Proprietary
Cloud Backbone Comparison – IP/MPLS vs.
Ciena’s Dynamic Converged Optical Ethernet Large Carrier Cloud Economic Analysis
53%
Less
Point-to-Point
On-Demand
Any-to-Any CBE
CBE
Common
Packet-
Optical Mesh
CBE
No
rma
lize
d N
ew
Eq
uip
me
nt
Cap
ex
to
20
15
IP/MPLS
Greenfield
IP/MPLS
Brownfield
Dynamic COE
Greenfield
Beyond the Network…
www.ipanematech.com
Beyond the Network…
www.ipanematech.com
Beyond MPLS Optimizing Performance in the Cloud
2
5
www.ipanematech.com
Headquartered in Paris, France
Globally, our systems support over 100,000 sites
Leading “Visionary” in Gartner WOC Magic Quadrant
Market leader for application aware network services
26
Ipanema Overview
www.ipanematech.com
Networks move from MPLS
to hybrid MPLS + Internet
Internet applications move from
recreational to business tools
Standard applications move to SaaS
In-house applications and resources
move to IaaS/PaaS
Unified Communications and video
become critical business applications
Users access applications from
everywhere: branch, home, hotel…
27
The Cloud Changes Network Requirements
MPLS
Cloud
Internet
www.ipanematech.com 28
Networks Must Support New and Traditional Business Applications!
Internal business applications (SAP, Oracle, homegrown…)
Financial consolidation and reporting (Oracle Hyperion, SAP Cartesis…)
SaaS collaboration (Office 365, Google Apps, LotusLive…)
Telepresence (Cisco, Polycom…)
SaaS CRM, ERP (SalesForce, Oracle On Demand…)
HighDef video delivery
(eLearning, Digital Signage…)
Unified Communications (MS Lync, Cisco UC…)
Social media as business tools (Facebook, Twitter, YouTube)
MPLS / Ethernet
Cloud Computing
Internet
www.ipanematech.com 29
Increasing Demands on Branch Locations The New Branch Communication Hub for Global Organizations
The changing traffic matrix
Networks are moving from point to point to any-to-any
Collaborative apps have multiple flows to each branch
Hosted virtual desktops do not “require” routing from the
desktop to the data center
Tremendous demand on the access and
distribution of data to and from branch locations
Business continuity requires making real-time
decisions & adjustments to session connectivity
Dynamically manage and control competition
between critical and less-critical applications
MPLS is inadequate
Service providers own only 1-2 access points
in a meshed, cloud network
CoS cannot address multiple critical application
traffic streams
Stateless QoS will be inadequate; per destination
QoS with fairness is required
Application SLAs are required
SaaS
IaaS/PaaS
International
Branch
Domestic Branch
Private DataCenter
Ethernet
MPLS
Internet Main site HQ
Private DataCenter
Public Internet
Cloud
Computing
www.ipanematech.com
The network is more business critical than ever
30
The Result Cloud computing shifts application delivery complexity to the network
Cloud Impact IT Network
Application
Delivery
Complexity
www.ipanematech.com 31
The Challenge
How do you cost optimize your WAN?
Take advantage of the Internet as a business network
Get global control without deploying technology everywhere
Reduce your WAN bandwidth requirements now and plan for tomorrow
Discover applications and their resources usage
Understand the root cause of application slowdowns
Communicate clear data about application performance
How do you get full visibility over your global network?
How do you guarantee application performance?
Control private and public applications over a hybrid network
Ensure excellent application performance to your distributed workforce
Manage social media and recreational applications
www.ipanematech.com 32
The Ipanema Autonomic Networking System (ANS
TM)
Autonomic
Global, distributed coordination between appliances guarantees performance
“Sense and Respond" dynamically adapts to traffic and network changes
Full control requires only 10-20% of sites with an appliance
Tightly coupled Application Visibility, QoS & Control, WAN Optimization and Dynamic
WAN Selection
Optimizes data transfers (CIFS...), interactive flows (Citrix...) and real-time flows
(Videoconference...)
All-in-one
Service Framework
Unified management GUI for all features
Objective based control enabling global WAN Governance
Multi-tenant platform (SALSA) that scales up to 10MM‟s users & 100K‟s sites
www.ipanematech.com
Guaranteed performance for business critical applications based on customer specific SLAs
Complete “autonomic” application management, visibility and control across an enterprise network
based upon global user defined objectives
Dynamic traffic optimization (session-by-session) across hybrid networks; both public or private
Control and optimization with absolutely no decisions or actions required; zero demands placed upon
IT staff resources
Multitenant Central Management System that scales to very large networks and large numbers of
customer domains
The Autonomic Network System Delivers
Enterprise Branch
SaaS
Data Center Service Provider
Data Center
Enterprise Data Center
Business
VPN
Internet @
33
www.ipanematech.com 34
Guaranteed Performance Over Any Topology
Autonomic Control Agent (ACA) Real-time Cooperation among ACA
One to Any
Tele-optimized Branch Offices
Single Data Center
Any to Any
Branch Offices
w/ inter-site traffic
Multiple Data Centers
Some to Many
Tele-optimized Branch Offices
Multiple Data Centers
www.ipanematech.com 35
Dynamic WAN Selection Guarantees Application Performance Across Hybrid Networks
Appliances cooperate to analyze available
bandwidth and quality for each path
Automatic path selection based on
bandwidth and congestion
Flows are automatically adjusted to
guarantee performance according to
individual SLAs
Dynamically selecting the network to match
performance objectives for each application
Paths are instantly adjusted to adapt to any
change in network and traffic conditions
Significant Benefits
Use broadband links in branches
Eliminate complex policy based routing
Increased reliability with dual
operators/networks
Datacenter #1
SAP
Datacenter #2
Sharepoint
Internet MPLS
Branch Office
XaaS
www.ipanematech.com
Complete visibility and control of
application performance
Global view of application usage and
performance across the enterprise network
Application SLAs are provisioned and
analyzed through clear KPIs (AQS, MOS)
Global application performance objectives
applicable throughout the entire network
Multi-tenant capability provides separation
between customer domains
Proven scalability, tested to 400,000+ sites
Customer web-based portal
KPIs, reports, analysis
Real time view of data
Help desk access
iPhone application
36
SALSATM
Scalable Application Level Service Architecture
Enterprise Applications
Applications Criticality
SAP
IP Telephony
Telepresence
CRM/Citrix
CIFS
Salesforce
Gmail
Sharepoint
Skype
YouTube
Top
Top
High
High
Med.
Med.
Med.
Med.
Low
Low
www.ipanematech.com
Fast to deploy
Network Transparency
Fully transparent at Layers 2 and 3
Fully compatible with Classes of Service
Provisioning
Install plug & play appliances in a few key locations
Use pre-defined application performance objectives
Get meaningful KPIs in a few days (MOS, AQS)
Then one click to activate control and optimization
over the global network
Easy to operate
Zero day-to-day adjustments: ANS™ automatically controls application
performance across the network
Application performance objectives are automatically enforced for new sites
Set SLAs for new applications and automatically enforce them over your
global network
You need only a few minutes per month to check application SLAs thanks
to simple KPIs
37
Fast to Deploy, Easy to Operate
www.ipanematech.com
Business Continuity
Guaranteed performance for critical business applications; now and as you
implement Cloud based solutions
Raw Performance
Improve application response times 10X or more
Bandwidth management
Delay bandwidth upgrades by 24 months
Divide by 3 the cost to transfer a Gbyte of data across the network
Service Provider Benefits
Get away from MPLS commoditization and margin erosion
Increase VPN profitability
Reduce churn and acquire new customers by selling value!
Deliver highly differentiated cloud services enabled by a Cloud-ready network
38
The Bottom Line
Have Superior WAN Optimization and Management
Have Superior Bandwidth Management
Move Beyond Dated MPLS
Q&A Session
Mitchell H. Auster Senior Advisor
Market Development
Ciena
Sarah Wallace Analyst
Heavy Reading
David White Senior VP Global
Business Development
Ipanema Technologies
Thank you for attending!
Upcoming Light Reading Webinars
www.lightreading.com/webinars.asp