Performance Objectives and Performance Components Chapter 2 Performance Objectives and Performance Components 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

From Chapter 1 Step 1: Learning Top-Level Requirements Many enterprise network designs fail because the executives involved have no clear-cut definition of success, and the objectives begin to vary, then thrash into chaos, during deployment. As a network architect, your most important job is to articulate the real requirements. 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

From Chapter 1 Step 2: Application Topology and Naming First, of course, you need to identify the endpoints, which are users and servers. Second, as you collect this information, introduce a naming structure to organize what you collect. Should also collect information on the computer- hardware type operating systems networking software type and version Knowing the system and networking software gives you insight into latency 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

From Chapter 1 Step 3: Addressing, Routing, and Switching Create a working cloud that interconnects your users and meets their needs. The backbone provides connectivity among the user endpoints. Steps 1 and 2, deal with a user-oriented worldview. Inside the network the requirements for efficient and reliable transmission are paramount. Remember your high-level goal is cost-effectiveness. 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

From Chapter 1 Step 4: Software After you have decided on the routing structure and where to use switches to optimize performance, you need to select software functions to enable on the routers and switches. For routers, a major software decision is the mechanism(s) they will use to find paths. Addressing heavily interacts with software selection. If you choose to use unregistered addresses at the edge of the network, but use the global Internet for at least some of your connectivity, you need either to tunnel or translate some of those addresses. 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

From Chapter 1 Step 5: Hardware and Media Speeds and feeds Speeds of media May or may not need switching on all interfaces 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

From Chapter 1 Step 6: Deployment Avoid doing massive cutovers, During times that real users are on the network. In hierarchical networks, it is best to start in one area, then add the core, and then add other areas. 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

From Chapter 1 Phase 1: Test network capability Phase 2: Initial trials with live data Phase 3: Just before full cutover Reflect workloads close to real environment Cut back over if necessary Phase 4: Production-level implementation Look for problems under a full load. 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

Chapter 2 Process of defining an internetwork that meets user performance requirements is iterative: Quantify user performance needs: Steps 1 & 2:define endpoints to be connected Trace the path that the end-to-end user flow will take. Step 3:address, routing and switching design. 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

Requirements may exceed available budget. Define internal constituents: Median and interconnection devices Requirements may exceed available budget. Chapter 2 focuses on Step 2-Application Topology and Naming 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

Flows Unidirectional Traffic (One direction only) Flows-unidirectional associations between source and one or more destinations. 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

Application Response Time Sum of measurements for a set of one-way transfers. Non-interactive:one-way transfer. Throughput and availability Interactive:become difficult when human perceptual factors become involved. Have response time Local processing in the client before first communications sent Queuing for transmission at the server’s communication interface Network transfer from the client to the server Processing in the server Network transfer from the server to the client Queuing in the client or gateway used by the client 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

Application Response Time Responsible for true network delay: Local processing in the client before first communications sent. Queuing for transmission at the server’s communication interface. Two unidirectional data relationships, one in each direction, for interactive applications. 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

Categorizing Flows Flows give ideas on the workload they will impose in a network. Basic factors that establish throughput: Sustainable avg rate Peak rate Approximate burst size 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

Categorizing Flows RFC 2430 PASTE categorize flows as: Network control: routing updates, SNMP, etc. Priority: App traffic sensitive to jitter(sensitivity to delay). Best effort: Data traffic that can be retransmitted, or one-way voice or video transmission such as radio or tv. 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

Categorizing Flows Flows can be set up: Implicitly Explicitly 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

Performance Values User-perceived performance depends on the sum of the hop-by-hop unidirectional paths in the network plus host processing delays 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

Measuring Workload ANSI X3.141-statistical considerations in measuring a workload in a network. User traffic peak 60-90 minutes after official work time. Peak load could occur in night with backups, data distribution, database updates. Busy hour;bimodal distribution. Morning and afternoon. 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

IETF Benchmarking Methodology Working Group Vendors specsmanship, smoke and mirrors. Comparable data from different vendors with which to evaluate these devices. 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

One way flow consists of some set of relays and outgoing links. Need to determine whether dynamic routing is sufficient to find the path needed to meet the service requirement, or whether explicit traffic engineering is needed. 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

Traffic Engineering and Queuing Policies Traffice Engineering: Explicitly assigning traffic to paths other than those that would be selected by ordinary dynamic routing. Queuing Policies Handling of flows in routers and switches. If output port is busy, traffic destined for it either is dropped or buffered. When port becomes available, a scheduler selects the queue from which it sends the next packet 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

Queuing Policies (cont) Paths inside routers are governed by queuing policies Links between routers are described with media specifications such as bandwidth and delay Basic delays-sum of the link and relay delays in the end-to-end paths 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

Flows can be: Point-to-multipoint Point-to-point 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

Refining User-Oriented Performance Edge of network At the user device there are two basic kinds of application-significant performance Response time for interactive apps Throughput for non-interactive apps 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

X3.102 Model Technology-independent User oriented Basis of national and international standards 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

X3.102 Three phases Access: connection setup, other preparation such as flow requests. User information transfer: actual communication. Disengagement: process of disconnecting interface and making ready for new connection. 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

Within each of these phases are: Successful performance, measured in times. Misperformance, measured as a probability. Nonperformance, measured as a probability. 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

Measurements taken go into a matrix of user-oriented performance parameters. 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

X3.102 introduces the idea of interface responsibility. If the source host is ready to transmit data through the network interface, it gives responsibility to the outgoing interface to accept the data. Measurement of communications takes place while the channel between the source and the destination hosts is responsible for the interface. 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

X3.102 introduces the idea of interface responsibility(cont) If the source stops being ready to transmit, or the destination stops being ready to receive, the interface responsibility changes. Must be a clear aggreement on network performance measurement and the contribution of hosts to user-visible performance, the network provider blames the user for poor performance and vice versa. 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

Special Medium Considerations Performance analysis deals with complex topologies by treating them as sets of point-to-point media. Demand Media that require real or virtual call setup at Layers 1, 2, or 3 are essentially unicast. Demand services have different requirements at each end of the spectrum. 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

Special Medium Considerations Analog or ISDN low-bandwidth links Extremely high bandwith virtual circuit services such as ATM. 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

Special Medium Considerations Analog or ISDN low-bandwidth Make sure network management and routing protocols do not trigger unecessary/expensive calls. Do not use routing protocols with periodic updates;likely to hold the link up indefinitely. If using RIP and IGRP make sure they don’t propagate over the dial up Be cautious using any routing protocol over dialup. They use keepalives. OSPF and RIP(1 or 2) with demand support minimize the costs of calls. 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

ISDN setup speeds are sufficiently fast that the incoming packets sent by the user, are stored in router RAM buffer until the demand circuit is ready to carry them out of the router. 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

ATM Even faster setup times. 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

Multicast Applications Modern routing protocols such as OSPF and EIGRP use multicasting.(to transmit a message to multiple recipents at the same time). MBONE Multicast tunnels running over general IP networks. Bandwidth intensive. Two multicast classes: Multipoint-to-multipoint Point-to-multipoint 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

Host Issues Many network performance problems are due to host, not network, capacity. End-to-end Sliding window designed to optimize throughput RFC 1323 Traffic Generation Computer/workstation traffic generated Oversubscribed media-uanble to carry traffic High-end Workstations and Servers- 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

Host Issues Broadcast and Multicast effects Processing load placed on hosts Traffic replication at Ports and Trunks Best way to avoid broadcast storms is to have protocols that avoid them. Hosts-CPU loading called broadcast radiation Host has to process every broadcast and at least examine each multicast. 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

Availability Fail to be available: Operational costs No connectivity Too slow Operational costs Redundant mechanisms Personnel Two paths-share load. If one path fails user notices drop in performance but not failure 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

Availability Failure Scope of Downtime Edge Connectivity WAN link fails High Availability to General Internet Endpoint multihoming-apps running on specific servers visible to the internet. High Availability to Arbitrary Internet server 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

Availability High Availability to Selected Partners-extranets Extranets use tunnels-software construct that carries payload information in a payload protocol over a delivery system. Appears to end system as point-to-point connection. 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

Availability VPN IP tunnels can interconnect members of a given VPN. Performance cannot be controlled because it tunnels through ISP’s. 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

Backbone Need to consider what reliability level is needed from the backbone. Use redundancy where it will help. Excessive use can increase convergence. Fault tolerance to be considered at the router and physical levels. 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing

Service-Level Agreements Excellent tools for requirements in the design level process. Involve host application performance as well as the network performance. Workload estimates: Application Overall environment Consider availability. Expectations of significant periods of downtime Terms of mean time between(total) failures and mean time to repair. 11/29/2018 Designing Routing and Switching Architectures. Copyright 1999 Macmillan Technical Publishing