ysis. The user can select a range within the text and then gets a word-cloud for that text range. They can then move the range through the text and ob- serve the changes, which means that the word-clouds have to be generated very quickly. While the text itself is static and thus would be amenable to using pre- calculation of the word-clouds using the existing, slower algorithms, the user is provided with a range of parameters to tune what exactly they see in the word-cloud. The range of parameters make pre-calculation unfeasible, it is thus necessary to have a fast word-cloud algorithm that generates stable word-clouds. In this paper we present a novel algorithm for fast generation of coordinated word-clouds in which all words have a completely fixed location. The remainder of the paper is structured as follows: Section 2 discusses previous approaches to creating comparable word-clouds, Section 3 provides an overview over the data we are working with, Section 4 presents our algorithm for stable word-clouds, which is evaluated in Section 5 and in Section 6 we conclude with future work.
An alternative is to build the reference corpora while the user browses the web. As the proxy tokenises and analyses each document anyway, we can keep track of the document frequency of each observed term. In this way, we can operate on a corpus that also reﬂects the user’s interests. As in the beginning such a continuously extended corpus is very small considering the number of terms seen so far, the ﬁrst few browsed documents will be analysed on a basis of pretty distorted document frequencies for the contained words. Accordingly, also the TF-IDF values will be distorted. Hence, a conversion into document wordclouds makes sense only after the corpus has reached a certain size. Empirically we found out, that already a corpus of around 3,000 to 4,000 unique terms was suﬃcient to obtain reasonable results in document visualisation.
Several improvements and extensions to wordclouds have been proposed in the last couple of years. Many of them address layout issues. For instance, Kaser and Lemire  present methods to reduce and balance the white space in HTML-based wordclouds. Seifert et al.  also propose space-ﬁlling word cloud algorithms, but use a different layout strategy to cope with convex polygons as boundaries. Yet another layout strategy is applied in the popular word cloud generator Wordle , which has been adapted with some modiﬁcations in the works on ManyWordle  and Rolled-out Wordles .
Based on the survey comments from students and our reflections as practitioners and researchers, there are some recommendations that could improve this particular assignment, as well as other assignments involving wordclouds. For example, one student explained, “I don’t really understand wordclouds, so I didn’t really understand the assignment.” It is important to clearly articulate the nature of wordclouds in general, and the origins of the particular word cloud, before proceeding with an assignment. It is also recommended educators engage students in word cloud analyses as a whole group prior to incorporating it as an individual assignment, as a few students reported difficulty. As one student shares, “It was difficult to analyze just the words without a context of sentences.” However, it should be noted that the analysis of wordclouds is, and should be, challenging. One student explained, “The word cloud doesn’t really give a good context for an analysis of the speeches and makes it quite a bit harder to figure out what is being talked about. I think that the only reason I was able to ‘see’ what was being talked about was that something just clicked in my brain.” Finally, attention should be focused on the word cloud with regards to the location and placement of the words. If the software allows, certain words should be kept together in order to preserve certain phrases crucial for understanding.
The support for multiple source formats is part of the effort to make MuWoC a versatile and useful visual analytics tool for textual data. However, since texts are the main intended form of input, there has to be a way to distinguish documents from different categories. These categories are the base for the creation of multiple wordclouds. Without the information about the belonging to categories, there would only be one cloud. To create separate ones, a support vector machine[VC74] (SVM), could be trained to classify the documents. This would take a training set, and manual labor for an initial classification. Since there already are other tools - for instance WEKA [HFH + 09] and RapidMiner [MWK + 06] - to classify texts, MuWoC does not implement such a feature.
or alphabetical order. Although they are proliferating in vi- sualization and text analysis, it is difficult for them to tell how the words related to a certain topic vary from those re- lated to other topics, since these words scatter in the wordclouds. Users would have to visually search for their desired words in the wordclouds for comparison, thus introducing additional overhead. However, creating semantic-preserving wordclouds is challenging. We have attempted to create context-preserving wordclouds with a force-directed algo- rithm [ CWL ∗ 10 ], but this approach has several drawbacks. First, the resulting wordclouds may not be stable; a slight change of the input words may result in very different wordclouds. Second, the created wordclouds may have very ir- regular shapes. In comparative visualization, wordclouds with regular shapes, such as rectangles, are preferred by users. Finally, the original semantic relations among words may be destroyed in the resulting wordclouds, which may confuse or mislead users in visualization.
Collins et al.  presented Parallel Tag Clouds, a method that aims to make comparisons easier by representing the documents as lists. Although alphabetical lists are informa- tive and easy to understand, our work aims to retain the aesthetic appeal of wordclouds while improving their infor- mativeness. The closest work to ours is Cui et al. , which was later improved by Wu et al. . This work proposes using a sequence of wordclouds along with a trend chart to show the evolution of a corpus over time. They present a new layout algorithm with the goal of keeping semanti- cally similar words close to each other in each cloud. This is a diﬀerent goal from ours: Preserving semantic relations between words within a cloud is diﬀerent than coordinating similarities across clouds, and does not necessarily result in similar documents being represented by similar clouds.
Having described the parameters used in the set up of the LES it is now time to pay attention to the most important parameters. As we are studying the mean profiles and the way they transport energy through the ABL this dictates which parameters we will change. Obviously the mean profiles itself are of great importance, as a change in the initial profile is likely to impact the results from the LES. In figure 3.1(a) the mean (liquid water) potential temperature profile is sketched. There are four distinct parts. In the first 500 m a well mixed layer is supposed, where the potential temperature stays the same, the mixing is due to turbulence. Between 500 and 1500 m the potential temperature raises due to the formation of clouds. Then between 1500 and 2000 m we find a strong 3 The geostrophic wind is the horizontal wind for which the Coriolis acceleration balances the horizontal pressure
The primary benefit of moving to clouds is application scalability. Unlike grids, scalability of cloud resources allows real-time provisioning of resources to meet application requirements at runtime or prior to execution. The elastic nature of clouds facilitates changing of resource quantities and characteristics to vary at runtime, thus dynamically scaling up when there is a greater need for additional resources and scaling down when the demand is low. This enables workflow management systems to readily meet quality-of-service (QoS) requirements of applications, as opposed to the traditional approach that required advance reservation of resources in global multi-user grid environments. With most cloud computing services coming from large commercial organizations, service- level agreements (SLAs) have been an important concern to both the service providers and consumers. Due to competitions within emerging service providers, greater care is being taken in designing SLAs that seek to offer (a) better QoS guarantees to customers and (b) clear terms for compensation in the event of violation. This allows workflow management systems to provide better end-to-end guarantees when meeting the service requirements of users by mapping them to service providers based on characteristics of SLAs. Econom- ically motivated, commercial cloud providers strive to provide better services guarantees compared to grid service providers. Cloud providers also take advantage of economies of scale, providing compute, storage, and bandwidth resources at substantially lower costs. Thus utilizing public cloud services could be economical and a cheaper alternative (or add-on) to the more expensive dedicated resources. One of the benefits of using virtualized resources for
few times 10 7 years (Blitz & Shu 1980; Larson 1981). The short lifetimes of molecular clouds are directly indicated by the fact that the range in age of the young stars associated with them is only about 10 to 20 Myr, comparable to the internal dynamical timescales of large molecular clouds (Larson 1981); age spans of this order are found both for the T Tauri stars in dark clouds (Cohen & Kuhi 1979) and for the subgroups of OB stars in more massive star-forming complexes (Blaauw 1964, 1991). Subgroups of OB stars whose ages are greater than 10 Myr no longer contain significant amounts of gas or dust, so the gas from which such groups form is evidently cleared away in a time of this order (Blaauw 1991); those clouds or parts of clouds that form star clusters are also dispersed in a similar time, since little or no molecular gas is seen within 25 pc of clusters older than 10 Myr (Bash, Green, & Peters 1977; Leisawitz, Bash, & Thaddeus 1989). There cannot be any long ‘dead time’ between the formation of a massive molecular cloud and the onset of star formation in it, since very few of the largest molecular clouds are not forming stars (Blitz 1991), and the number of molecular clouds of all masses that are not forming stars is only about compa- rable to the number of star-forming clouds (Mooney & Solomon 1988; Solomon, this conference). Therefore the total lifetimes of molecular clouds, or at least of those parts of them that form systems of stars, cannot exceed twice the period of active star formation in them, or perhaps ∼ 20 Myr (see also Elmegreen 1991a, who estimates a cloud lifetime of ∼ 40 Myr from similar arguments). Thus there is no need, as has often been supposed, for molecular clouds to be supported against gravity in a quasi-equilibrium configuration for many dynamical times. While it is true that virial balance is often roughly satisfied, at least in the larger clouds (Larson 1981), this does not rule out the possibility of rapid dynamical evolution of these clouds, or of those parts of them that form groups or clusters of stars.
This section expands on the reference architecture described earlier in this document and rationalizes the architecture with data artifacts described in the Use Cases and Interactions for Managing Clouds white paper ( DSP-IS0103 ). Familiarity with DSP-IS0103 is required to fully understand this section’s contents. By understanding this section, the reader can expect to come to an understanding of the architectural interactions and functional capabilities that would be needed by an offering from a cloud service provider. The information contained here should be considered referential only and non-normative.
Aslam, Mudassar and Gehrmann, Christian and Björkman, Mats (2012) Security and Trust Preserving VM Migrations in Public Clouds . In: The 2nd IEEE International Symposium on Trust and Security in Cloud Computing, in conjunction with IEEE TrustCom-12, 25-27 June 2012, Liverpool, UK.
Figure 4 shows the execution time for FSaaS application on desktop, public cloud and two private clouds. Experiments confirm with the followings. Firstly, enterprise portability is achieved and the FSaaS application can be executed on different platforms. Secondly, the improved FSaaS application can go for 100,000 simulations in one go on Clouds. Although above 100,000 simulations in one go, factors such as performance and stability must be bal- anced, before tuning up the capabilities of our FSaaS. The six-core processing rack server has the most advanced CPU, disk, memory, 64-bit operating system and networking hardware, and is not surprising that it is always the quick- est. Although the desktop has similar hardware specification to server, it comes out slowest in all experiments. The difference between the Public Cloud (large instance) and Private Cloud (virtual server) is minimal. Although the large instance of a public cloud has the edge on hardware specification against the Virtual Pri- vate Cloud (VPC), the networking speed within the VPC is faster than the Public Cloud, and this explains the small differences between them.
II. D OMAIN S PECIFIC L ANGUAGES AND C LOUDS A DSL is a programming language or an executable specification language that offers, through appropriate notations and abstractions, expressive power focused on, and usually restricted to, a particular problem domain . DSLs have been used in many domains, particularly due to their expressiveness, runtime efficiency and reliability due to their narrow focus. More recently, DSLs for clouds have been proposed for high performance computing  business process management  and business applications . Data cloud specific DSLs, such as Pig Latin from the Apache Pig project, are employed for analyzing large data sets .