CODA performance in the real world

The most ambitious implementation of the Jefferson Lab data acquisition system (CODA) to date is for the CLAS spectrometer in Experimental Hall B. CLAS has over 40,000 instrumented channels and uses up to 30 front-end (FASTBUS/VME) crates in the DAQ subsystem. During the initial experiments we found that performance of the fully instrumented DAQ system did not scale as expected based on single point to point benchmarks. Over the past year we have been able to study various performance bottlenecks in the CLAS DAQ system including front-end real time performance, switched 100BaseT Ethernet data transport, and online data distribution and recording. Performance tuning was necessary for components on both real time (VxWorks) and UNIX (Solaris) operating systems. In addition, a new efficient Event Transfer System (ET) was developed to provide faster online monitoring while having minimal impact on data throughput to storage. We discuss these issues and efforts to overcome the real world problems associated with running a high performance DAQ system on a variety of commercial hardware and software