Campus Design Principals

Transcription

1 Campus Design Principals John Hicks Internet2

2 Overview Campus design challenges Edge Networks Layer2 design Core Networks Research and EducaEon Networks Network Management

3 Campus Design Challenges Many are not structured properly and can t effecevely uelize high bandwidth REN conneceons Many make heavy use of NAT and firewalls that limit performance Many are built with unmanaged network equipment that provide no ability for monitoring or tuning the network

4 How Best to support R & E Research and EducaEon needs flexible and open networks Things to consider NAT makes some things hard (H.323 video conferencing) Filtering makes it hard for researchers, teachers, and students to do intereseng things Your campus network must not be the bozleneck Make a plan for improvement without a plan, how will you get there

5 Campus Design Rules Minimize number of network devices in any path Use standard solueons for common situaeons Build Separate Core and Edge Networks Provide services near the core Separate border routers from core Provide opportuniees to firewall and shape network traffic

6 Core Versus Edge Core network is the core of your network Needs to have reliable power and air condieoning May have muleple cores Always route in the core Edge is toward the edges of your network Provide service inside of individual buildings to individual computers Always switch at the edge

7 Core versus Edge (cont.) There is usually a correspondence between building separaeon and subnet separaeon Switching inside a building RouEng between buildings This will depend on the size of the network Very small networks can get by with doing switching between buildings Very large networks might need to do roueng inside buildings

8 Layer 2 Network Design Guidelines Always connect hierarchically If there are muleple switches in a building, use an aggregaeon switch Locate the aggregaeon switch close to the building entry point (e.g. fiber panel) Locate edge switches close to users (e.g. one per floor) Max length for Cat5 is 100 meters

9 More Layer2 Guidelines Minimize the distance between elements Never daisy chain

10 Start Small

11 Build Incrementally

12 Grow Hierarchy

13 Add Redundancy

14 Core Network RouEng versus Switching Layer 2 versus Layer 3 Routers provide more isolaeon between devices (they stop broadcasts) RouEng is more complicated, but also more sophisecated and can make more efficient use of the network, parecularly if there are redundancy elements such as loops

15 Layer 3 Switches Many vendors use the term Layer 3 Switch. These are contradictory terms Layer 3 = RouEng Switch = Layer 2 What vendors mean is that it is a device that can be configured as a router or a switch or possibly both at the same Eme

16 Core Network Reliability is the key Remember many users and possibly your whole network relies on the core May have one or more network core locaeons Core locaeon must have reliable power UPS bazery backup (redundant UPS as your network evolves) Generator Grounding and bonding Core locaeon must have reliable air condieoning

17 Core Network At the core of your network should be routers you must route, not switch Routers give isolaeon between subnets

18 Simple Core Network

19 Where to put Firewalls Security devices must be placed inline This means that the speed of the firewall affects access to the outside world Try to have parts of your network non- firewalled This will allow full bandwidth, un- filtered access to the Internet

20 Where to put Servers? Servers should be on a high speed interface off of your core router Servers should be at your core locaeon where there is good power air condieoning

21 Border Router Connects to outside world RENs and Peering are the reason you need them Must get Provider Independent IP address space to really make this work right

22 Layer 2 and 3 Summary Route in the core Switch at the edge Build star networks don t daisy chain Buy only managed switches re- purpose your old unmanaged switches for labs

23 No Network is Perfect What happens to your network when you get a 100Gbps conneceon from Internet2? Where are the bozlenecks? How will you improve performance? What is your plan? Can you upgrade without disrupeng the whole campus?

24 Research and EducaEon Networks High bandwidth networks 10G, 40G and 100G backbones Research typically needs uncongested networks Which means many RENs are lightly used with lots of unused capacity (we call it headroom) Low latency Terrestrial fiber Open Networks with no filtering Firewalls can make it hard for ad- hoc aceviees

25 Why a REN? Enable research or services that could not be accomplished otherwise Cost Savings (buyers club) Aggregate demand from muleple parees Vision of building alliances Successful RENs find that there are unanecipated benefits

26 Why Are We Doing This? Our goal is to build networking capacity to support Research and EducaEon Remember: University = Research & EducaEon Buying all service from your local ISP is a losing game you will spend more money and not have control of the network The pazern around the world is to build regional, naeonal, and larger Research and EducaEon Networks (RENs)

27 REN versus Campus Network The Campus Network is the foundaeon for all Research and EducaEon acevity Without a good campus network, the Research and EducaEon Network can t work as well as it should The campus network is the foundaeon that the REN is built upon

28 REN Ecosystem A layered model Global ConnecEvity Regional RENs NaEonal Research and EducaEon Networks All users are connected at the campus network level No scienest is connected directly to a NaEonal Network. They are all connected to campus or enterprise networks

29 Network Management Keeping Track Of StaEsEcs For purposes of accouneng and metering Faults (Intrusion DetecEon) DetecEon of issues, TroubleshooEng issues and tracking their history TickeEng systems are good at this Help Desks are a useful to criecal component

30 ExpectaEons A network in operaeon needs to be monitored in order to: Deliver projected SLAs (Service Level Agreements) SLAs depend on policy What does your management expect? What do your users expect? What do your customers expect? What does the rest of the Internet expect? What s good enough? % UpEme? There's no such thing as 100% upeme

31 Baselining What is normal for your network? If you ve never measured or monitored your network you will need to know things like: Typical load on links Level of jizer between endpoints Typical percent usage of resources Typical amounts of noise : Network scans Dropped data Reported errors or failures

32 Why do all this? Know when to upgrade Is your bandwidth usage too high? Where is your traffic going? Do you need to get a faster line, or more providers? - Is the equipment too old? Keep an audit trace of changes Record all changes Makes it easier to find cause of problems due to upgrades and configuraeon changes Keep a history of your network operaeons Using a Ecket system lets you keep a history of events. Allows you to defend yourself and verify what happened

33 Why manage Networks? AccounEng Track usage of resources Bill customers according to usage Know when you have problems Stay ahead of your users! Makes you look good. Monitoring soqware can generate Eckets and auto- maecally noefy staff of issues. Trends All of this informaeon can be used to view trends across your network. This is part of baselining, capacity planning and azack deteceon.

34 Special thanks to NSCR for slide material and organizaeon GRNOC, Indiana University

35 QuesEons?