Network Performance

• 3/4/2009

Quality of Service

• Want speed!
  • Get up to 10x slower
• Want everything
  • Get up to 10% of traffic as high priority
• don't forget, one "class" of service is none/best effort!
• Quality of Service for packet networks is the assurance that certain special packets will encounter fewer problems on the network than all the other packets.

QoS Bandwidth Broker

QoS Bandwidth Broker

QoS Policy Server

QoS Policy Server

QoS Compared

Performance Metrics (review)

• latency
• jitter

Performance Metrics (review)

• throughput (not quite bandwidth)
• packet loss
  
• device queue depth: ignore latency if any loss

Network Application Types

• client-server
• peer-to-peer
• traffic distribution, network utilization and architecture
  • Network traffic distribution follows a power law; the rule of thumb is that a power law has an 80%-20% split. So 80% is one thing, and everything else is 20%. Before p2p and Google as a homepage, 80% of network traffic stayed within a building.

Network Application Types

• bulk transfer
  • bandwidth
• real-time
  • jitter
• interactive
  • latency: 400 ms, 100 ms, 80 ms

Network Application Types

• LAN or WAN
• point-to-point or multi-point
• simplex or full duplex
  • caching, conferencing or casting
  • store-and-forward or real-time
  • exhibition or participation
  • application sharing or collaboration

Hiding Performance Problems

• jitter
  • delay equalization buffer
• packet loss
  • redundancy
  • audio: white noise <16ms
  • video: previous frame

Latency

• It's the Latency, Stupid is another view of why latency is the killer.

Performance for What?

• Nagle's Algorithm is good for the network and bulk transfer, but bad for interactivity.

Simple Performance Tests

• ping www.example.org
• traceroute www.example.org
• nmap -p 80 www.example.org
• Internet Traffic Report
• How does ping stack up for the metrics?

Advanced Performance Tests

• netperf, iperf, or nuttcp
• NDT
• What does NDT add to performance testing?

Performance Testing Types

• SNMP/RMON (mainly errors)
  • ifInErrors and ifOutErrors
  • ifInOctets and ifOutOctets
  • RMON Statistics
• observation (netflow, or agent on end system)
• synthetic

Troubleshooting Example: IPv6

• disable IPv6
  • Windows
  • Linux
    • SUSE
    • Debian
      • Ubuntu
  • Mac OS X

Example: SNMP

• Manning Hall has some thin WAPs on ports 2, 4, 6, 8, 10, 12, 14 of 172.29.209.113
• look at, for instance, ctEtherManualConfigEntry.prof, IfTable.prof, and RmonStatsTable.prof (I browsed /opt/1q/lib/profiles)
• getTable -n ${ROSTR}12 -s 172.29.209.113 -prof IfTable
• aliasGrep -n ${ROSTR}12 -s 172.29.209.113 -active

Practical Troubleshooting

• Troubleshooting definition

1. check likely failure modes first
   • what changed recently? what's new?
2. check things that are very easy or very cheap to check
   • replace the cable with a known good cable
3. trace the power
   • is it plugged in?
4. ask a question with a known answer
   • can I ping it?

Example: IPv6

• "The Internet is slow"
• need to have a baseline ... NDT
• look for a firewall
• replace cables
• Why the Internet is slow laugh

Projects

example

Finally ...

ENJOY SPRING BREAK!