**4.4 Analysis**

The next step in the data processing chain is then the analysis. There are two types of analysis: a scheduled analysis organized centrally and then the end-user, so-called chaotic analysis. Since processing of the end-user analysis jobs often brings some problems like a high memory consumption (see Figure 12) or unstable code, the scheduled analysis is organized in the form of so-called analysis trains (see [52]). The trains absorb up to 30 different analysis tasks running in succession with one data set read and with a very well controlled environment. This helps to consolidate the end-user analysis.

The computing model assumes that the scheduled analysis will be performed at Tier-1 sites, while the chaotic analysis and simulation jobs will be performed at Tier-2s. The experience gained during the numerous Data Challenges, the excellent network performance, the stable and mature Grid middleware deployed over all sites and the conditions at the time of the real data taking in 2010/2011 progressively replaced the original hierarchical scenario by a more "symmetric" model often referred to as the "cloud model".
