5. VOIP Access / Service Enablement – The ADC Need
• The peering, trunking, and consumer access points are
governed by SBC-s, P/I-CSCF-s, Soft Switches and media
gateways, all of which have limited scaling
– New calls per second < 100
– Concurrent sessions < 10,000
• Main Scaling Challenges
– Application state synchronization for billing and policing
– Failures and performance bottlenecks
– Provisioning OPEX and ineffective over subscription
Slide 5
6. Deployment Scenarios – Application/Service Tier
• SIP applications include
– Traditional VoIP
– Conferencing / Video Conferencing
– Voicemail / IVR
– Presence
– Messaging
– Call Centers
– Interactive E-learning
– And many more…
• Additional Scaling Challenges:
– Concurrent streaming sessions
– Bandwidth and processing requirements for new Codecs and Video
standards
– Sophisticated application logic and call flows
Slide 6
Interactive Gaming
Collaboration–Virtual
Whiteboarding
Video
C
onferencing
Instant
M
essaging
Presence
Distance Learning
MixedMedia
Communications
Personal Mobility
VoIP
SIP
IP
Convergence
7. Application / Service Tier – The ADC Need
The overall scale challenge:
• Scale-as-you-Grow
• protecting previous investment and guaranteeing service
quality
Slide 7
9. TDM-like scaling
• “TDM over IP” Architecture
– Scaling in the traditional TDM architecture was normally
achieved by statically configuring sets of clients to a specific
server
– Complex Static Provisioning
– Inefficient Over-Subscription models
• Inefficient and not agile
Slide 9
10. IP Scaling Advantages vs. TDM
• IP Architecture – Dynamic Load Balancing
– The same utilization advantages afforded by the flexibility of IP
packets switching (vs., circuit switching) can be incorporated
into the scaling model
– Servers are all configured identically and clustered into a pool,
abstracted as a service available to all clients
• New servers are added dynamically on-demand
• Scaling is linear and isolated from the provisioning process
– Servers may have different capabilities
• Server selection identifies the required service for each
transaction for optimal server selection
• Huge cost saving – up to 50% over the TDM-like model
Slide 10
11. Load Balancing Concepts – Cluster Management
• Load Distribution
– New client sessions or messages are distributed by the ADC amongst
cluster service through a selection of distribution algorithm:
• Weighted round robin – Short transactions (messaging / presence / DNS)
• Least number of users / calls – Video transactions
• Lowest CPU utilization – Complex application transaction
• Transaction based switching
– The ADC inspects the message content to apply the distribution across
multiple server pools according to:
• Codec type
• Device User-Agent
• Content URI
Slide 11
SIP Server Farms
SIP: INVITESIP: INVITE
12. Load Balancing Concepts – Cluster Distribution
• Persistent transactions
– Some applications require that the same server will handle multiple sessions
that form a transaction
• Multiple callers for a Video conferencing service
• User’s registration and incoming/outgoing transactions
• User’s portal access and media transactions
– Persistency Types include
• Call-ID
• Call-ID + Conference-ID
• User ID
• Source IP
• Any L7 Parameter
Slide 12
Call-ID = 1223023@a.com
Server 1
Invite SIP: bob@b.com
From : alice@a.com
To: bob@b.com
Call-ID: 1223023@a.com
SIP Proxies
13. Load Balancing Concepts – Cluster Distribution
• Persistent transactions
– Some applications require that the same server will handle multiple sessions
that form a transaction
• Multiple callers for a Video conferencing service
• User’s registration and incoming/outgoing transactions
• User’s portal access and media transactions
– Persistency Types include
• Call-ID
• Call-ID + Conference-ID
• User ID
• Source IP
• Any L7 Parameter
Slide 13
Call-ID = 1223023@a.com
Server 1
BYE
From : alice@a.com
To: bob@b.com
Call-ID: 1223023@a.com
SIP Proxies
14. Advanced Scaling Features
• Transport Conversion
– TCP TLS UDP / IPv4 IPv6
• Call Admission Control
– Centralized CAC
– Burst handling and Priority call guarantee
• Header Manipulation
– Normalizing message format
– SIP and SDP headers
• Geographic Load Balancing
– Disaster Recovery – uninterrupted Failover and Failback
– Scaling to a Multi-Site service
Slide 14
16. Summary
Adapt your architecture to advanced IP delivery models
• Achieve Linear Scale in your cost model
• Protect your investment
• Simplify application logic
• Reduce your time-to-market
Slide 16