Changes between Version 54 and Version 55 of Check


Ignore:
Timestamp:
Oct 29, 2010, 1:10:44 PM (14 years ago)
Author:
alexandru.olteanu
Comment:

--

Legend:

Unmodified
Added
Removed
Modified
  • Check

    v54 v55  
    413413=== Performance ===
    414414
    415 The lag between the detection of a node failing and restoring the service, by turning it on for another node, is usually around a few miliseconds. We have tried a few measurements, for both local islands and remote islands. In these examples, the Self-healing service is running in the same network as the Bucharest island and we introduce node failures in both islands.
     415The lag between the detection of a node failing and restoring the service, by turning it on for another node, is usually around 30-40 miliseconds. We have tried a few measurements, for both local islands and remote islands. In these examples, the Self-healing service is running in the same network as the Bucharest island and we introduce node failures in both islands.
    416416
    417417[[Image(sh-performance.png, center, 600px)]]
     
    419419The data from the above screenshot is better shown in this table:
    420420
    421 ||= Test no. =||= Node failure =||= Detection time =||= Service restored time =||= Lag =||
    422 || 1 || remote || || ||
    423 || 2 || remote || || ||
    424 || 3 || remote || || ||
    425 || 4 || remote || || ||
    426 
    427 You may notice that the mean restore time for local failures is ~ and the mean restore time for remote failures is ~.
     421||= Test no. =||= Node failure =||= Detection time [s] =||= Service restored time [s] =||= Lag [ms] =||
     422|| 1 || remote || 24.182 || 24.297 || 115 ||
     423|| 2 || local || 30.741 || 30.775 || 34 ||
     424|| 3 || remote || 04.876 || 05.016 || 140 ||
     425|| 4 || local || 37.247 || 37.279 || 32 ||
     426
     427You may notice that the mean restore time for local failures is ~30-40 ms and the mean restore time for remote failures is ~100-150 ms. The difference between remote and local times is most probably caused by network lag times and remote times can vary even more depending on the connection between the computer running the Self-healing service and the network island.
    428428
    429429= Proposed Check Demonstration Sketch =