Challenges of Big Data Analytics in High-Energy Physics

Challenges of Big Data Analytics: volume, variety, velocity and veracity
Screenshot of CERN Big Data Analytics presentation

There are four key issues to overcome if you want to tame Big Data: volume (quantity of data), variety (different forms of data), velocity (how fast the data is generated and processed) and veracity (variation in quality of data). You have to be able to deal with lots and lots, of all kinds of data, moving really quickly.

That is why Big Data Analytics has a huge impact on how we plan CERN’s overall technology strategy as well as specific strategies for High-Energy Physics analysis. We want to profit from our data investment and extract the knowledge. This has to be done in a proactive, predictive and intelligent way.

The following presentation shows you how we use Big Data Analytics to improve the operation of the Large Hardron Collider.

Displaying Dimuon Events from the CMS Detector using D3.js

Physicists working on the CMS Detector
Physicists working on the CMS Detector

I became a Python geek and GnuPlot maniac since I joined CERN around three years ago. I have to admit, however, that I really enjoy the flexibility of D3.js, and its capability to render histograms directly in the web browser.

D3 is a JavaScript library for manipulating documents based on data. This library helps you to bring data to life leveraging HTML, CSS and SVG, and embed it in your website.

The following example loads a CSV file, which includes 10,000 dimuon events (i.e. events containing two muons) from the CMS detector, and displays the distribution of the invariant mass M (in GeV, in bins of size 0.1 GeV):

Feel free to download the sample CSV dataset here.

Further reading: D3 Cookbook

CERN: Where Big Bang Theory meets Big Data Analytics

Screenshot of SQL Plan Baselines with Oracle Enterprise Manager at CERN
Screenshot of SQL Plan Baselines with Oracle Enterprise Manager at CERN

The volume, variety, velocity and veracity of data generated by the LHC experiments at CERN continue to reach unprecedented levels: some 22 petabyte of data this year, after throwing away 99% of what is recorded by the LHC detectors. This phenomenal growth means that not only must we understand Big Data in order to decipher the information that really counts, but we also must understand the opportunities of what we can achieve with Big Data Analytics.

The raw data from the experiments is stored in structured files (using CERN’s ROOT Framework), which are better suited to physics analysis. Transactional relational databases (Oracle 11g with Real Application Clusters) store metadata information that is used to manage that raw data. For metadata residing on the Oracle Database, Oracle TimesTen serves as an in-memory cache database. The raw data is analysed on PROOF (Parallel ROOT Facility) clusters. Hadoop Distributed File System (HDFS), however, is used to store the monitoring data.

Just as in the CERN example, there are some significant trends in Big Data Analytics:

  • Descriptive Analytics, such as standard business reports, dashboards and data visualization, have been widely used for some time, and are the core applications of traditional Business Intelligence. This ad hoc analysis looks at the static past and reveal what has occurred. One recent trend, however, is to include the findings from Predictive Analytics, such as forecasts of sales on the dashboard.
  • Predictive Analytics identify trends, spot weaknesses or determine conditions for making decisions about the future. The methods for Predictive Analytics such as machine learning, predictive modeling, text mining, neural networks and statistical analysis have existed for some time. Software products such as SAS Enterprise Miner have made these methods much easier to use.
  • Discovery Analytics is the ability to analyse new data sources. This creates additional opportunities for insights and is especially important for organizations with massive amounts of various data.
  • Prescriptive Analytics suggests what to do and can identify optimal solutions, often for the allocation of scarce resources. Prescriptive Analytics has been researched at CERN for a long time but is now finding wider use in practice.
  • Semantic Analytics suggests what you are looking for and provides a richer response, bringing some human level into Analytics that we have not necessarily been getting out of raw data streams before.

As these trends bear fruit, new ecosystems and markets are being created for broad cross-enterprise Big Data Analytics. Use cases like the CERN’s LHC experiments provide us with greater insight into how important Big Data Analytics is in the scientific community as well as to businesses.

CERN: The world’s first website went online 20 years ago today

CERN website dispayed in Line Mode Browser
CERN website dispayed in Line Mode Browser

On this day 20 years ago the world’s first website went live. The website, created by Tim Berners-Lee at CERN, was a basic text page with hyperlinks and went live on August 6, 1991.

The website was hosted on Berners-Lees’ NeXT computer, the first web server ever, which had a note taped to the front that said: “This machine is a server. DO NOT POWER DOWN”.

NeXT computer used as first World Wide Web server
NeXT computer used as first World Wide Web server

Today this computer is displayed in the CERN Computer Center, which is just located next to my office.

[Update 30 Apr 2013]: CERN is bringing the very first website back to life at its original URL. If you’d like to see it, point your browser to: http://info.cern.ch/hypertext/WWW/TheProject.html

Data Science: Enabling Research at CERN with Big Data

Wow, time flies. One year has passed since I started to work at CERN as a data scientist. CERN, surrounded by snow-capped mountains and Lake Geneva, is known for its particle accelerator Large Hadron Collider (LHC) and its adventure in search of the Higgs boson. Underneath the research there is an tremendous amount of data that are analysed by data scientists.

Filters, known as High Level Triggers, reduce the flow of data from a petabyte (PB) a second to a gigabyte per second, which is then transferred from the detectors to the LHC Computing Grid. Once there, the data is stored on about 50PB of tape storage and 20PB of disk storage. The disks are managed as a cloud service (Hadoop), on which up to two millions of tasks are performed every day.

High Level Trigger data flow
High Level Trigger data flow, as applied in the ALICE experiment

CERN relies on software engineers and data scientists to streamline the management and operation of its particle accelerator. It is crucial for research to allow real-time analysis. Data extractions need to remain scalable and predictive. Machine learning is applied to identify new correlations between variables (LHC data and external data) that were not previously connected.

So what is coming up next? Scalability remains a very important area, as the CERN’s data will continue to grow exponentially. However, the role of data scientists goes much further. We need to transfer knowledge throughout the organisation and enable a data-driven culture. In addition, we need to evaluate and incorporate new innovative technologies for data analysis that are appropriate for our use cases.