The CMS trigger: taming the firehose

Much like pressing a button on your phone to choose which pictures you take, the CMS trigger makes decisions about which collision events to save. The decision is made in a fraction of a second, reducing the potential data stream out of the CMS detector by 99.999 percent.

Much like pressing a button on your phone to choose which pictures you take, the CMS trigger makes decisions about which collision events to save. The decision is made in a fraction of a second, reducing the potential data stream out of the CMS detector by 99.999 percent.

Most of us have a fairly sophisticated particle detector in our pocket: a phone with one or more cameras. To a physicist, each camera sensor is a dense photon detector, sending its readout to your phone’s local storage in the form of images. As the sensor itself draws very little power, one sure way to never miss a moment worthy of capture is to have your phone’s camera continuously record images. The approach isn’t without problems: You’d quickly run out of storage space on your phone and most of the pictures would be of limited value: pocket lint, tabletops, floors, and plain ol’ darkness. Fortunately, your phone has a simple way to avoid this pitfall: a trigger mechanism, namely, you pressing a button when you see that perfect sunset. Analogously, the CMS detector is a large digital camera, and CMS physicists have built an elaborate trigger system into the detector to avoid saving too many particle physics equivalents of pocket lint.

In normal operation, the LHC’s beams of accelerated protons cross each other 40 million times a second, resulting in collision events the CMS detector could record. Given the typical size of a recorded event coming out of CMS, saving every crossing’s worth of data would involve approximately 12 terabytes (a terabyte is a thousand gigabytes – likely a bit larger than your laptop’s hard drive) every second, over 40,000 terabytes an hour. To put that into perspective, the more than 1 billion users of Facebook upload “just” over 20 terabytes of photos every hour. Fortunately, the types of particle events that CMS physicists want to study are rare. For example, Higgs bosons are produced in about one out of every 10 billion collisions. The CMS trigger system is designed to identify potentially interesting events in near-real time and chooses to save only those.

The CMS trigger consists of two stages. The first, known simply as the L1 trigger, consists of custom electronics that read the output from each subdetector within CMS. The L1 then compares those inputs to a set of thresholds for each subdetector and decides whether to keep the information from that collision or throw it away. This entire decision-making chain occurs in less than 4 millionths of a second. Every second, its decisions take the 40 million crossings and reduce them to around 100,000 to consider further.

The remaining data travel to the second stage, known as the high-level trigger (or HLT). The HLT consists of several thousand computers working in concert. For each event, it takes the bits of information from individual subdetectors and reconstructs a more complete picture of the particle interactions that occurred. Thanks to the L1’s work, the HLT gets a relatively luxurious 100 thousandths of a second to contemplate each event, throwing out a further 99 percent of events. The events that survive the HLT (about 400 every second – or 0.001 percent of the collisions delivered by the LHC), are used to obtain every CMS physics result. A much more detailed description of the CMS trigger system and its performance during the first run of the LHC can be found in a paper recently submitted to the Journal of Instrumentation.