Plans for measurements
----------------------

0. Objectives of the measurement
  A. Decide if final goal is feasible
  B. Find optimal operating parameters
  C. Identify slow components that need further optimization

I.  Measure throughput
  A. Setup options
    1. Fixed/variable event size
    2. RU-BU transfer: Ethernet/GM/barrel shifter/mem copy/zero copy
    3. EVM in same host as RU/BU/FLT
    4. RU in normal host/ in Dell
    5. concurrent processes on CPU/PCI
    6. Trigger source (TTCrx event by event, with DMA, fedkit...)
    7. Random trigger generator
    8. Simulate multiple RU/BU messaging
    9. Replace GM with MAZE
   10. Replace Myrinet with Ethernet

  B. Scan parameters
    1. Event size (mean and RMS)
    2. Block sizes (FedKit, network)
    3. Message bundling (all messages)
    4. Number of resources
    5. Number of blocks in FedKit
    6. Backpressure levels

II. Measure where time is spent
  A. Separate CPU usage by layer
    1. shortcut layers
  B. Find slow part of code inside given layer
    1. Internal clock in processor (Emilio?)
    2. Dummy PCI accesses recorded by analyser
  C. Use a profiler tool
    1. oprofile.sourceforge.net