文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > A Simulated Fault Injection Tool for Dependable VoD Application Design
侵权投诉

A Simulated Fault Injection Tool for Dependable VoD Application Design

This work presents a simulation-based tool for dependability-oriented design of Video on Demand (VoD) applications. The tool is organized in a layered architecture, so that simulation models can be built and detailed according to a hierarchical and modular

UnitsValue orDescription

Rangeλnet[1/h][0.01-100]Fault rate for the Network LinksΤnet[h]0.1Fault duration for the NetworkLinks

λS[1/h][0.05-50]Fault rate for the Film Servers(FSs)

µS[h][0.00025-0.25]Fault mean duration for the FSsσS[h][0.0000025-0.0025]

Standard deviation for the FSsλds

[1/h]

[10-9-10-6]

Fault rate for the Video Server(for a single memory location)

Table 1 – Fault injection parameters

The following fault models have been assumed:

Corruption of data and control packets – Packets

are sent over the network. If the Network is hit by afault while a packet is traversing it, the packetbecomes unusable, regardless of the number of bit-flips. As far as video is concerned, fault effects are adirect consequence of the MPEG compression

technique. If an I frame gets corrupted, the

Player will be unable to decode the entire Group OfPictures (GOP), i.e., the current sequence of I, P, andB frames. As a consequence, a glitch will occur in thevisualization process. The duration of this glitch willbe (at most) 1-2 seconds. If a P frame gets corrupted,the player will be unable to decode the currentsequence of B frames comprised between the latest Ior P frame successfully received and the corrupted Pframe. If a B frame gets corrupted, only that oneframe is lost. As far as audio is concerned, corruptionof a frame results in that frame being played outpoorly. Faults in the Network Link disappear after apredefined amount of time, as reported in Table 1. Adifferent scenario stems from the corruption ofcontrol packets. In this case, the corruptedsynchronization message is simply retransmitted.

Corruption of entries in the User List and File Listdatabases – These data structures are stored in theVideo Server RAM memory. If the memory is hit bya fault, the affected database entry becomesunusable, regardless of the number of memory flips inthe entry. If this happens in the User List, it results ina Hard Service Failure (HSF), i.e. the user is deniedaccess to any document. If this happens in the MediaList, it results in a Soft Service Failure (SSF), i.e. theClient is denied access to the requested document, butnot to other documents. We assumed that after afailure has occurred, the memory location whichoriginated the failure is healed. As a consequence,faults in the data structures are transient, but theirduration is not pre-defined.

Servers’ downtime – We assumed Film Servers aretaken down, according to an exponential distribution,for some predefined time duration, in order toperform maintenance operations. If a transmission isin course, recovery must be initiated. Fault injectionparameter values are reported in Table 1. Performance parameters (such as the effective linkcapacity, decoding time, and disk I/O time) forsimulation setup were derived from a trace file collectedon the field. Since the demo took place on Sundaymorning, external traffic was very low and the connectionhad an effective bandwidth of almost 2 Mb/sec. Rangesfor video decoding times of different MPEG files arereported in Table 2. Values refer to a Sparc 20. PentiumII machines were 2-3 times faster.FRAME TYPE

MIN [s]

MAX [s]

I0.0060.245P0.0130.096B

0.005

A Simulated Fault Injection Tool for Dependable VoD Application Design

0.077

Table 2 - Video decoding times (Sparc 20)The arrival of requests from users has been modeledaccording to an exponential law. The simulation wasdriven using real video and audio files (the same used inthe actual system).

4. Simulation Results

In this section, we present a portfolio of dependabilitymeasures obtained from simulation experiments on theVoD system described in section 2. These resultsdemonstrate only some of the potentialities provided byour simulation framework.

4.1 Mean Time Between Packet Loss

We monitored five kinds of events: corruption of I, P,and B frames, corruption of audio packets, and corruptionof synchronization messages.

Figure 6 - Mean Time Between Packet Loss(MTBPL)

上一页第5页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email