adam's notes

Common Network Performance Issues

Bursty data transfer is when the sequence in which the packets are delivered and the variable intervals between packets arriving do not materially affect the application.

  • used in file transfer

Quality-of-Service (QoS) protocols and appliances are designed to support real-time services.

  • voice and video applications carry real-time data
    • have high bandwidth requirements
      • typically less of a problem than packet loss, latency, and jitter

Bandwidth

Bandwidth is the amount of information that can be transmitted.

  • measured in bits per second (bps), or some multiple thereof
  • expresses the available capacity of the link
  • when monitoring, need to:
    • distinguish between the nominal data link/Ethernet bit rate
    • throughput of a link at layer 3
    • the goodput available to an application
  • bandwidth for audio depends on:
    • sampling frequency (Hertz)
    • bit depth of each sample
    • e.g., early digital telecommunications links were based on 64 Kbps channels
      • derived from calculating:
        • voice frequency range is 4,000 Hz
          • must be sampled at twice the rate to ensure an accurate representation of the original analog waveform
        • sample size is 1 byte (8 bits)
          • 8 KHz x 8 bits = 64 Kbps
  • VoIP bandwidth requirements for voice calling can vary
    • allowing 100 Kbps per call upstream and downstream is sufficient in most cases
  • Video bandwidth is determined by:
    • image resolution (number of pixels)
    • color depth
    • frame rate (fps)

Bottlenecks

A bottleneck is a point of poor performance that reduces the productivity of the whole network.

  • may occur because:
    • a device is underpowered or faulty
    • user or application behavior
  • identifying cause of bottleneck:
    • identify where and when the network overutilization or excessive errors occur
    • if problem is continual,
      • likely to be device related
    • if problem occurs at certain times,
      • likely to be user or application related

Packet Loss

  • packet loss is expected
    • but only to a degree
  • larger network = more packet loss during heavy traffic
  • identifying packet loss:
    • run a packet sniffer on affected segment
    • high numbers of TCP retransmission and duplicate acknowledgements are strong indicators
    • know where and when packet loss occurs to find device that is dropping frames
  • reasons for dropped packets:
    • server, router, or switch is overloaded
    • power outage
    • firewall is blocking packets from a known destination
    • malicious actor is interfering with network transmissions
    • faulty firmware is causing packet processing errors

Latency and Jitter

Latency is the time it takes for a transmission to reach the recipient.

  • measured in milliseconds (ms)
  • can test latency of a link with
    • ping
    • pathping
    • mtr

Jitter is variation in the time it takes for a signal to reach the recipient.

  • manifests as an inconsistent rate of packet delivery
  • measured in milliseconds (ms)
    • using an algorithm to calculate the value from a sample of transmit times
  • can test jitter with mtr

Latency and jitter deal with problems of timing and sequence of packet delivery.

  • not significant problems when data transfer is bursty
  • real-time applications are more sensitive
    • manifests as echo, delay and video slow down
    • if packets are delayed, arrive out of sequence, or are lost
      • receiving host must buffer received packets until the delayed packets are received
      • if packet loss is so excessive that buffer is exhausted,
        • then noticeable audio or video problems (artifacts) are experienced
  • when assessing latency, consider Round Trip Time (RTT)
    • VoIP requires
      • RTT < 300 ms
      • jitter < 30 ms
      • packet loss < 1%
  • with Wireshark
    • gives each packet a timestamp relevant to when the last frame was sent from the very first transmitted frame
      • can find delays in TCP conversation during a session
    • can plot a sequence graph to visually represent how this delay behaves
      • line should have a gradual, steady increase upward to the right
      • optimal network should have small gaps between each transmission
      • the more longer and jagged the graph, the more latency is introduced

Graph View

Backlinks