A Gigabit/s Data Acquisition System Andrei Sukhanov, Piotr Kulinich, Pradeep Sarin Abstract- A fully pipelined data acquisition system has been developed for the PHOBOS experiment at RHIC in BNL. The system began operation in 1998; after the most recent upgrade in 2004, it reached a trigger rate of 575 Hz, processing speed of 200 MB/s, and maintained a sustained data-sinking rate to HPSS of 50 MB/s. The system receives data from 135,168 silicon pad detectors (12-bit amplitudes) and from 2,000 scintillating detectors (time and amplitudes). The data from the silicon detectors are transmitted through two 120 MB/s fiber links to a 24-node processing system (Mercury Computer System, Inc.) embedded into a VME crate and interconnected using a RACEWAY switching fabric. Each node (PPC750 300 MHz) compresses a part of the data stream using lossless Huffman algorithm. The four-fold compressed data are sent directly from the nodes to a model TK200 data recorder (Conduant Corporation) through an FPDP link. The event building is done on the fly by adjusting offsets and lengths of a scatter- gather DMA transfer. The TK200 has a disk capacity of 3.2 Tbyte and aggregate bandwidth of 200 MB/s, and is connected to the HPSS over a single Gigabit fiber connection. Under laboratory conditions, the TK200 was able to handle 120 MB/s of recording and 70 MB/s of reading simultaneously. The upgraded system was used for data taking during RHIC run 2005 and recorded over 0.5 billion (50 Tbyte of data) of heavy-ion collisions over a period of two months. he data acquisition system (DAQ) at PHOBOS [1] receives data from 135,168 silicon detectors and 2000 scintillation detectors. It has following components : • Processing farm with 24 processing nodes (CE) with MPC750 (300 MHz processor, 1 MB of L2 cache SRAM, 32 MB of DRAM). The farm also contains 2 input ports (RINT) and one output port (ROUT). The nodes and ports are located on 8 VME motherboards (PCH6), all connected to a RACEway switching fabric [2][3][4]. • Model TK200 data recorder with disk capacity of 3.2 Tbyte [5] connected to the output port of the farm and over Gigabit interface to a central storages system (HPSS). • Several programmable logic modules based on a custom VME board PCDP [6] for triggering and synchronization. • RS485 link, also based on PCDP, for taking data from FASTBUS and VME crates with ADCs and TDCs. • Trigger monitor: MVME-2306 board with PPC 604 processor, VxWorks OS • Host CPU (Themis USPIIe with UltraSPARC 400 MHz CPU) for monitoring and code development for the farm. Fig1. Structure of the P HOBOS Data Acquisition System. The whole system (excluding TK200 data recorder) occupies a single VME crate (DAQ crate). The signals from the silicon sensors are digitized in Front End Controllers (FEC), which are located in the experimental area [7]. Each sensor is digitized using a 12-bit ADC but samples are stored as 16-bit words. The multiplexed data are transferred to the PHOBOS T