Simulation of the LAr felix slice

After the meeting here it was decided to simulate the felix network LAr slice for PhaseI.

Simulation development

We identified the following features that need to be added to the simulation:

  1. [generator] Host multi-puerto: los felix van a tener al menos 2 puertos de salida (uno a cada switch). Ahora el template del pdm solo permite 0 o 1 puerto.
    status: [DONE]
  2. [generator+simulation] multi-path + bonded links: los datos de un felix hacia un swrod puede tomar 2 posibles caminos (uno por cada switch). Ademas, Los 2 links que salen del felix en la realidad son bonded-links (la aplicación lo ve como uno solo).
    status: [DONE] . Details: Routing for multi-path flows
  3. [simulation] Host multi-flow: los felix van a generar datos y otros flujos (ej: monitoring). Ahora la simulacion permite un unico flow por generador. (esto por ahora no es tan prioritario, porque vamos a arrancar unicamente con datos).
    status: [WIP]

Some extra data:

  1. [simulation] initialization time: con ~40 modelos el tiempo de inicialización se fue a 106s (crece rapido, con 6 modelos estaba en <10s). Esto ya lo teniamos en el radar y por la lectura de parametros de configuracion de Scilab. Es poco prioritario porque el workaround es mirar facebook mientras inicializa, y porque arreglarlo no es trivial.
  2. [simulation] execution time: generando paquetes a 1MHz (con paquetes de 1K son 1Gbps) en 15 servers tarda 335s en simular 1s. No hice mas pruebas, el objetivo es simular a 40Gbps, que tardaría 3.7hs en simular 1s.
    No hay mucho para hacer al respecto... mas que terminar mi doctorado con un modelo híbrido smile
  3. [PowerDEVS] GUI: la interfaz de powerDEVS es inusable con tantos modelos. No es una traba y se puede usar el visualizador de ONET.

Felix servers parameterization (traffic characterization)

  1. Traffic characterization:
    1. To characterize the felix traffic patter we will follow up with Jorn.
      [DONE]. See meeting Caracterizando la generacion de trafico de los servidores felix (Jorn, Matias, Andy)
    2. Implement felix traffic into the simulation
    3. The monitoring traffic will be a percentage of the data traffic.
    4. Still need to understad DCS traffic.
  2. simulation scenarios: it is interesting to use the simulation in the following scenarios (with this priority):
    1. [WIP] Base case: felix servers and swrod only with 1:1 data flow communication. There should be no bottleneck or queueing. Boring case.
    2. Add Monitoring traffic: add the monitoring traffic on top of the data. see traffic characterization below. ¿how does it affect data traffic?¿how to setup queue priorities? estimatemax % of monitoring traffic according to link BW.
    3. Add DCS traffic: add the DCS traffic, which is required to have a minimum latency. ¿how to setup queue priorities to minimize lantency? Estimate expected DCS latency.

Validation

Before jumping into the simulation of bigger topologies, it would be good to perform validations of the simulation against some real system.

  1. Eukeni agreed to prepare a simple topology, maybe 4 hosts + 1 switch.
  2. We will help to prepare the applications to generate and measure traffic.
  3. The focus will be on testing the priority queues using several flows, and validating at the throughput of each flow.
  4. It would be nice if we could additionally validate latency (probably RTT) and queue occupancy, but it will depend on what we can extract from the switch.

-- MatiasAlejandroBonaventura - 2016-11-14

Edit | Attach | Watch | Print version | History: r3 < r2 < r1 | Backlinks | Raw View | WYSIWYG | More topic actions
Topic revision: r3 - 2016-11-17 - MatiasAlejandroBonaventura
 
    • Cern Search Icon Cern Search
    • TWiki Search Icon TWiki Search
    • Google Search Icon Google Search

    Main All webs login

This site is powered by the TWiki collaboration platform Powered by PerlCopyright &© 2008-2020 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding TWiki? Send feedback