Desirable properties of computer networks

1. General scope – It is preferable to have computer network designs that satisfy a wider range of requirements and support a wider range of hardware and software. Designs become less scalable as they become more specific to a particular problem or requirement.
2. Scalability – Computer networks should be able to handle small and large number of nodes without a significant reduction in efficiency or performance.
3. Robustness – The operation of a computer network should minimise the impact from problems and errors within the network and from the nodes.
   • Networks should assist administrators and users in the detection and identification of problems and errors.
   • Networks should have safety barriers to reduce the spread of problems from one part of the network to other parts of the network.
   • Networks should continue operating in a stable manner after a problem has been identified and removed or fixed.
   • Networks should maintain sufficient operating information to assist the analysis of nodes and systems that function incorrectly instead of ceasing to operate (Byzantine failure).
4. Automatic configuration – Configuration changes due to new or replacement hardware and software should require minimal administrative or user intervention.
5. Predictability – The operational characteristics of a computer network should be sufficiently deterministic such that similar designs should have similar operations. Predictable networks are easier to analyse and repair.
6. Incremental design – Incremental changes to parts of a computer network should be possible to allow the integration of new hardware or software.

Design issues common to networking devices

• Location of destinations – Layer 2 devices tend to move traffic between destinations that are directly connected. By comparison, Layer 3 devices tend to move traffic between destinations that are reachable, which often require intermediate connections.
• Availability of network topology information – Layer 2 devices tend to learn network topology information from immediate neighboring devices. By comparison, Layer 3 devices tend to acquire network topology information from a larger number of network devices. The cost of acquiring better topology information is the need for protocols to perform handshaking, update acknowledgements and modifications to data frames.
• Aggregation of forwarding headers – Layer 2 devices tend to have less aggregation, but, frame processing is less intensive and device configuration is less complex. Layer 3 devices tend to enable more aggregation, but, frame processing is more intensive and device configuration is more complex.
• Number of protocols to support – Supporting more protocols increase backwards compatibility, but, at the cost of more intensive frame processing and more complex configuration.
• Use of proprietary protocols or protocol extensions – Proprietary modifications can improve performance, but, at the cost of less compatibility with other networking devices.

Evolution of network design themes

Attributes of network services provided to users

Making decisions about the network services to provide requires a mix of the following attributes. It often is not practical or economical to maximise or optimise every attribute!

• capacity (bandwidth)
• latency (delay)
• reliability (probability of failure)
• maintainability (time to restore system after failure)
• availability (uptime)

These attributes may be defined in a service level agreement (SLA).

Relative comparison of latency and bandwidth

An improvement in latency often improves capacity, but, an improvement in capacity often makes latency worse.

Capacity (bandwidth):

• 10,000 Mb/s – 10 Gigabit Ethernet
• 1,600 MB/s – double data rate SDRAM
• 1,000 Mb/s – Gigabit Ethernet
• 100 Mb/s – Fast Ethernet
• 86 MB/s – SCSI magnetic disk drive

Latency (delay):

• 15 ns – Intel Pentium 4
• 52 ns – double data rate SDRAM
• 5.7 ms – SCSI magnetic disk drive
• 190 ms – 10 Gigabit Ethernet
• 340 ms – Gigabit Ethernet
• 500 ms – Fast Ethernet

Availability measured in allowed downtime per week

• 99% uptime is equivalent to 1.68 hour
• 99.9% uptime is equivalent to 10 minutes
• 99.99% uptime is equivalent to 1 minute
• 99.999% uptime is equivalent to 6 seconds

Client server network flow model

This is the most commonly recognised model since it is often found in both enterprise (internal to an organisation) and Internet (accessible to the public) applications. Flows tend to be directional and asymmetric, but, often with predictable patterns to the network flows.

Hierarchical client server network flow model

This model is becoming more widespread as more applications move to the cloud computing architecture. Flows tend to be directional and asymmetric, but, often with different patterns for each tier of the hierarchy.

Peer to peer network flow model

Nodes tend to operate at the same level of the network hierarchy and typically share a common service profile.

Distributed processing network flow model

This is a specialised model which often require high performance requirements from the network design.

Hierarchy and diversity are critical to network design

Hierarchy is the number of tiers of interconnection nodes and the amount of concentration of network traffic at the interconnection nodes. It determines the structure and scalability of the network.

Diversity is the amount of choices within each tier of the network. It affects the redundancy of the network.

Network design requires making trade-offs between hierarchy and diversity in order to achieve the system requirements.

Core/distribution/access architectural model

This architectural model partitions the network into 3 conceptual tiers:

1. Core tier
   • Very high speed throughput, usually at Layer 3
   • No policies that would slow traffic down, e.g. ACLs or filters
   • Provides redundancy and scalability
2. Distribution tier
   • Aggregation point for access switches, using both Layer 2 and Layer 3 switching
   • Apply policies for quality of service (QoS) and security
   • Provides high availability and load balancing
3. Access (edge) tier
   • Provides access to end-users and devices
   • High port density and low cost per switch port
   • Increase network convergence by supporting a wider range of devices and traffic

Key steps in network design

1. Collect information and generate system requirements
2. Develop the physical and logical architecture
3. Develop implementation and verification plans, including equipment provisions