Archana Ganapathi's research while affiliated with Splunk and other places

... Techniques have been proposed to identify and resolve data quality issues across multiple data sources, such as missing or inconsistent values [4]. Data quality issues pose a significant barrier to operationalizing big data and can lead to uncertainty and disruptions if not appropriately addressed [5]. ...

... Meanwhile, research communities in Visual Analytics and Human-Computer Interaction have investigated new techniques and systems that can help end-users in building ML models [18,48,54]. Such approaches are largely based on our theoretical understanding of how people explore information and gain insights using interactive systems, including sense-making loop [42], information foraging [41], and characterization of tasks and queries in Exploratory Visual Analysis [4,12,49]. Understanding how users use AutoML systems to perform their exploratory model building in constructing, understanding the resultant models, and recognizing the implications of their model choices is an important topic in visual analytic because the way in which a user builds a ML model is highly exploratory and can be directly supported by visual interfaces [16,31]. Gaining a deeper understanding of how users think with AutoML systems and where things may fail is crucial for determining how we can better support their ML model building process with visual analytic solutions. ...

... Actually, although real-world applications usually prefer row stores for transactions due to the given relational model, they may desire another specialized data format for analytical queries. For instance, complementary to column store, databases with novel data models are increasingly popular for analytical workloads, e.g., graph databases [15,33], spatial databases [76,173], and time series databases [24,140]. To absorb the advancements of these new databases, a well-customized HTAP system demands a re-design of data synchronization approaches, delta buffer, meta-data, execution engine, and query optimizer. ...

... Schema-on-read (Deutsch, 2013), (Mendelevitch, 2013), (Jacobsohn and Delurey, 2014) is an agile approach to data storage and retrieval that defers investments in data organization until production queries need to be run by working with data directly in native form. Schema-on-read functions have been implemented in a wide range of analytical systems including Hadoop (Hadoop Team, 2015), (Schau, 2015), Splunk (Bitincka et al., 2012), Apache Spark (Spark Team, 2015), Apache Flink (Markl, 2014), and even relational databases (Liu and Gawlick, 2015). It is also possible to use machine learning tools to extract schemas from source data (Yeh et al., 2013). ...

... New tools and systems that would help researchers share tools and results, and to validate previous results are now possible. Suggestions such as Dataforge [25], paralleling Sourceforge, and the use of cloud computing [26] are very good examples of approaches that would help researchers in experimental algorithmics and systems research. ...

... A large number of approaches have been proposed to isolate critical kernel code from less critical kernel extensions, as existing extension mechanisms were found to threaten system stability in case of misbehaving extensions [44], [45], [46], [47], [48], [49]. Similar to privilege separation, most work in this field has focused on how to establish isolation between the kernel and its extensions [50], [51], [52], [53], [54], [55], [56], [57], [58], [59], [60], [61], but only little work considers the problem of identifying what to isolate for achieving improved fault tolerance at an acceptable degree of performance degradation. ...

... They also discussed various traits of web analytics, its application areas (Burby and Brown, 2007). Ganapathi & Zhang showed in his paper the art of data summarization and its applications in managing systems effectively (Ganapathi & Zhang, 2011). Rapoza stated the web analytics from new perspectives and discussed innovative features of web analytics and its pros and cons with a new look (Rapoza, 2010). ...

... This may cause additional space overhead, which increase the infrastructure requirements and costs. Therefore, managing these mixed workloads (OLTP and OLAP) in the data management systems is a big challenge for current disk-based DBMSs [5]. ...

... Equipment failure prediction can be primarily classified into prediction methods based on mechanism models and data-driven prediction methods. Wang et al. [11] proposed a fault detection method for cloud-computing systems based on adaptive monitoring. By monitoring various attributes of the system, a correlation analysis was conducted to establish the correlation between each measure, and key measures were selected for system monitoring. ...

... Compression techniques in Hadoop delegate the computation load from I/O handlers to the processor. Compression improves both time efficiency and network traffic if I/O is the latency and network traffic bottleneck [6]. A set of configuration parameters in Hadoop control compression on a per-job basis. ...