The Network Dave DevereauxWeber University of WisconsinMadison Internet
The Network Dave Devereaux-Weber University of Wisconsin-Madison Internet 2 Commons Site Coordinator Training September 27, 2004 Austin, Texas
Wiring and Station Cables Common practice is unshielded twisted-pair (UTP) according to the specs: • CAT 3 [old] supports 10 Mbps Ethernet (10 base-T) • CAT 5 [modern] supports 10 base-T, 100 Mbps (100 base-TX) and 1000 Mbps (1000 base-T) Ethernet • CAT 6 [new] supports CAT 5 applications +
Wiring and Station Cables Actual wire used and quality of installation may vary widely – know your wiring! Important to consider the station cables • Don’t use sub-CAT 5 station cables for 100 Mbps connections. • Silver Satin telephone line cords are not CAT 5.
Ethernet LAN 10 / 1000 Mbps Full- and Half-Duplex • Half-duplex: send or receive, one at a time (listens for collision). • Full-duplex: send and receive simultaneously (does not listen for collision). • 10 Mbps Ethernet supports half-duplex; full-duplex is not consistently implemented. • 100 Mbps supports half- and full-duplex. • Modern devices can auto-sense speed and duplex.
LAN: Switches vs. Repeaters (hubs) are old technology. A repeater sends (repeats) packets that are incoming on one port, out all other ports (I know you’re out there somewhere!). Can only operate in half-duplex mode. Bandwidth and jitter provided to any single device is highly dependent on the LAN traffic.
LAN: Switches vs. Repeaters A switch learns the MAC addresses of the devices connected to it, and sends packets directly and only to the target end-point. Provides much more consistent bandwidth and latency (low jitter). A well-designed switched LAN is important for videoconferencing. Repeater-based LANs should be upgraded to switched for videoconferencing!
LAN: Ethernet Duplex Mismatch “One of the most common causes of performance issues on 10/100 Mb Ethernet links is when one port on the link is operating at half-duplex while the other port is operating at fullduplex. ” • http: //www. cisco. com/warp/public/473/3. html
LAN: Ethernet Duplex Mismatch “There is a silent performance-killer out there, one so inconspicuous that it is hardly ever looked for or even suspected. You could suffer from it and never know it, as it robs a site of performance but not connectivity. This performance-killer has a name: Ethernet duplex mismatch. ” • http: //www. hostingtech. com/nm/01_01_mismatch. html
LAN: Ethernet Duplex Mismatch If one end of a connection (device or Ethernet switch) is set for autonegotiation, and fails to see autonegotiation at the other end, the former sets itself to the default, half-duplex. Auto-negotiation can sometimes fail, even when both sides are set to auto (although this isn’t as prevalent as in the past).
LAN: Duplex Mismatch – Detection Microsoft Windows doesn’t display the auto-negotiated duplex setting. Some routers re-negotiate autospeed or auto-duplex, which can introduce jitter.
LAN: Ethernet Duplex Mismatch
LAN: Duplex Mismatch. Prevention Our recommendations: • Don’t use hubs for videoconferencing • If building wiring is sub-CAT 5, then set switch ports to 10/half • If building wiring is CAT 5 or better, then set switch ports and devices to 100/full if supported on switch.
Router Provides interface to the WAN. • Intranet, commercial Internet, and Internet 2 connections. • Typically, every networked device at an Internet 2 -connected institution has connectivity to Internet 2.
WAN Segments Service Speed Sample Uses T 1 1. 5 Mbps remote building; extension center DS 3 45 Mbps inter-campus; Internet (I 1) connection OC 3 155 Mbps inter-campus; I 1 & Internet 2 connection OC 12 622 Mbps I 1 backbones; Internet 2 connection OC 48 2. 4 Gbps Gigabit Etherne t 1 Gbps I 1 and Internet 2 backbones advanced inter-campus connections when have access to dark fiber
Indiana University Abilene NOC Weathermap
Traffic on the Network Typical university today: • IP – TCP – UDP • IPX [diminishing] • Appletalk [diminishing]
Traffic on the LAN Unicast : one-to-one Multicast: one-to-many Broadcast: one-to-every
Unicast Most common traffic Common applications: mail, Web browsing, file transfer, etc.
IP Multicast A one-to-many mode of transmission Network numbers 224. 0. 0. 0 through 239. 255 are reserved for multicast. Examples of multicast applications: • Vic/rat videoconferencing • Centralized PC software administration tools such as Symantec Ghost
IP Multicast – Leak Problems Beware: high rates of unpruned multicast can adversely affect videoconference performance. Use a network traffic and protocol analyzer to identify this problem.
Broadcast A one-to-every mode of transmission Used by network protocols including ARP and IPX, Net. BIOS system discovery, and name resolution. All devices on the network must process every broadcast packet; high broadcast rates can divert processing capacity. If the broadcast domain is too large or unusually active, the activity required at the end-point to deal with the broadcasts could diminish performance.
Broadcast A healthy network should have less than 100 broadcast packets per second. Check using a network traffic and protocol analyzer tool.
Firewalls A firewall is a network node that acts to enforce an access control policy between two networks, e. g. , between a university intranet and the commercial Internet. Used to secure IT resources against external attacks and break-ins. Network-layer firewalls typically make their decisions based upon port numbers and source/destination addresses. Application-layer firewalls act as proxies.
Firewalls H. 323 uses the IP ports: • Statically-assigned TCP ports 1718 – 1720 and 1731 for call setup and control. • Dynamically-assigned UDP ports in the range of 1024 – 65535 for video and audio data streams. Firewalls don’t allow unrestricted ports. Typical modern firewalls and H. 323 don’t get along so well.
Firewalls – Solutions for H. 323 [bad; non-scaleable] Allow unrestricted ports for specific, known, external IPaddresses. [better, but still not so good] Use feature of some videoconferencing clients to confine dynamic ports to a specific, narrow range. [OK, but extra admin work] Use Ridgeway Systems H. 323 application proxy. [best] Use a firewall that snoops on the H. 323 call set-up channels (static ports) and opens ports for the audio/video (dynamic ports) as needed.
NATs Allows multiple computers behind the NAT to share one external network address. Uses: • Alleviate shortage of IP addresses • Security – obscures view of the network from outside • Flexible network administration Not commonly used at universities on the campus level. Used somewhat in corporations. Common in small offices and at home – behind DSL, cable modem, or ISDN network service.
NATs Difficult to use H. 323 behind NATs. Some videoconferencing terminals provide features to work with NAT – refer to videoconferencing terminal documentation.
Latency is the time required for a packet to traverse a network from source to destination. Components of latency include: • Propagation delay: the time it takes to traverse the distance of the transmission line; controlled by the speed of light in the media; rule-of-thumb: 20 ms San Francisco to New York.
Latency Transmission delay: the time it takes for the source to put a packet on the network. Ruleof-thumb: < 1 ms. Store-and-forward delay: the cumulative length of time it takes the internetworking devices along the path to receive, process, and resend the packets. Rule-of-thumb: variable, and depends upon network load.
Latency Rule of thumb: • A one-way delay of: - 0 – 150 ms provides excellent interactivity 150 – 300 ms is OK 300 – 400 ms is bad 400+ ms is unacceptable
Jitter is variation in latency over time. If the endpoints are on switched LANs, then the primary source of jitter is variation in the store-and-forward time, resulting from network load. H. 323, particularly audio, is adversely affected by high levels of jitter. What is high? Rule of thumb?
Packet Loss Packet loss is typically due to congested links and routers. • 1% is noticeable • 5% becomes intolerable
Qo. S Not currently feasible on commercial Internet and Internet 2 networks for production, regular use. Internet 2 is working on Qo. S plans, but the current over-provisioned Internet 2 network doesn’t dictate need. Is useful on over-utilized intranet WAN links.
Qo. S How: • Some videoconferencing terminals can set the IP precedence bits. Use that for marking and priority queuing on the WAN. Or: • Use a H. 323 Proxy for consolidation of traffic to a single address, router access list for marking, and priority queuing on the WAN.
Qo. S Caution! • The wrong implementation could result in unwanted tradeoffs, e. g. , packet loss improves but jitter gets worse.
The End-to-End Performance Problem Scenario • Users on two different campuses of a university are experiencing poor video and audio in a conference. • Each user is supported by a different group of videoconferencing engineers. • Each campus is supported by a different group of network engineers. • The wide-area network is supported by a third group of network engineers.
The End-to-End Performance Problem • How do the users get timely, useful assistance? • How is network problem resolution coordinated?
The End-to-End Performance Problem Obstacles • Different groups, schedules, and priorities. • No one engineer has a complete understanding of the entire network path. • No one engineer can gain access to all the network nodes (routers, switches) along the path to inspect for trouble. • Communications are inconsistent from engineer to engineer.
The End-to-End Performance Problem Solutions • Articulate the E-2 -E problem to network management and engineers on all campuses. • Establish reliable communication tools, and insist that engineers utilize the tools. • Hold regular meetings; bring all engineers together in one place and time to share information. • Have good network documentation for all networks.
H. 323 is Network Sensitive! The big problems are: • Half/Full-duplex mismatches • Packet loss • Jitter • Substandard horizontal wiring or station cables • Multicast leaks • High broadcast rates
Recommendations Develop a close relationship with the network engineers and NOC. Make sure they understand what’s being done with videoconferencing and the network sensitivity of IP-based video. Articulate the End-to-End Performance Problem to network engineering and operations management. Champion ways to reduce the problem. Be sure to open trouble tickets with your NOC so that a problem history is maintained.
Recommendations Use switched Ethernet. Watch out for duplex mismatches. Keep an eye on utilization of WAN links, packet loss, and jitter. Make sure you don’t have broadcast or multicast leaking problems. Make sure wiring is up to the task.
Recommendations Have engineers in the videoconferencing support group trained to understand networking issues and tools.
The Network Information on these slides courtesy of Doug Pearson Indiana University And David Devereaux-Weber University of Wisconsin-Madison
- Slides: 44